diff --git a/ASE_exported/data/atek/cubercnn/train/81075/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81075/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..96a5bd3e1e3641a866678384817c3099fc1f97c0 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81075/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed79fa1b2085d6aa5cc0f7b477232dd682246a71e025f0db8c6e6ff899dd2903 +size 36538 diff --git a/ASE_exported/data/atek/cubercnn/train/81081/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81081/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..85b8419a59e51e5b808d96acc4d41c80bc8849eb --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81081/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5387d3fa257855bc4bebc603de08a3bea19bd6e9881cf7e4d64fcf2542b8e3af +size 14133 diff --git a/ASE_exported/data/atek/cubercnn/train/81247/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81247/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..a302a49e9370c284f87eb20a31ee83ef7d1ed0a1 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81247/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec1b39daf64553223c7b86e46ed09fb22500ba5ca7006cf1b55eb628689e9cef +size 29710 diff --git a/ASE_exported/data/atek/cubercnn/train/81411/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81411/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..5e92bebdc200f5a37addeff6bc5914f72d29887c --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81411/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e84301806da08c5d852fafe6cf407c747a4684f102ffc3e5e7275d986c324ba1 +size 38551 diff --git a/ASE_exported/data/atek/cubercnn/train/81623/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81623/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..bf375f9d4896ea21746c8589d966ec4e7559e229 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81623/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8296015f61e82c9d1fcdf41bdea4a0a94ddae6fc572a0195083f349f911513d +size 18851 diff --git a/ASE_exported/data/atek/cubercnn/train/81640/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81640/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..1ff950a95e5d37dcb33f26526d729640c36b2308 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81640/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca19c1dd112d602489fa901dd926035918e632d4d5bc7dc04c6e5f58eea5ef12 +size 17375 diff --git a/ASE_exported/data/atek/cubercnn/train/81849/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/81849/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..bb96016ccfc80d80d274cc23dbe1f489f23b844c --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/81849/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64e35f77e5b5591492f20fe3f3935f767067b8c82ce477a3e59c9727d44fbde7 +size 30881 diff --git a/ASE_exported/data/atek/cubercnn/train/82052/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82052/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..eb3be84af71567c3bc5472c4b79e6d98bbe55aa1 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82052/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2dc570c5ca6d047f23b56703a7f6f8c683c9ca2fd0ccc89daac7a9cd57ad6e64 +size 37310 diff --git a/ASE_exported/data/atek/cubercnn/train/82099/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82099/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..c552c61a7ab40db3ae5f04f367b25c06046b00e1 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82099/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b56d34a040bd6d99298b58a37bc0451c1d91de85439eed7c1a29b6b57c15d365 +size 19466 diff --git a/ASE_exported/data/atek/cubercnn/train/82260/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82260/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..2819838fb881f24751cf275a8e5038c563303aa4 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82260/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cccaa1055ffe86df203c531fdc507eabc9f26e4a09b2c2e97e323d603714db35 +size 26293 diff --git a/ASE_exported/data/atek/cubercnn/train/82294/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82294/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..b23911b3a0c6948c65f99d0c48230c0d0ba45caf --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82294/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e1a5808fa146bc63716f982d4864b02174026031c3d9f43fcb9436bbfd86e0c +size 52230 diff --git a/ASE_exported/data/atek/cubercnn/train/82436/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82436/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..12de7623708b1aa84820e026f19b0f02eb86b896 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82436/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be4a25c8efeff5de9de3d93b9b19b10bc99eb1dea412dd0c7a4ba2061f81141e +size 41001 diff --git a/ASE_exported/data/atek/cubercnn/train/82604/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/82604/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..09e3fc42e25d560fdb627bb23cee58a11a76db36 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/82604/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04915c1024276928c0c980376a52f359a491cc00f43c342185488e6940572251 +size 40031 diff --git a/ASE_exported/data/atek/cubercnn/train/83117/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83117/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..82c8f1c07d139c710fde339902c951a64ad954ae --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83117/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5c19230ee611940fcb98e5e47c9c3db042919cb2b7ae3e1cc5889778dfd64df +size 28670 diff --git a/ASE_exported/data/atek/cubercnn/train/83174/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83174/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..238164e1ccb644417e3f4cfdbc174c77ceced520 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83174/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:981426e85df47ff7e510fac4b1e1cfcc2d62c99cd29b4f4134ae867f3b1c3211 +size 4183 diff --git a/ASE_exported/data/atek/cubercnn/train/83180/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83180/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..1d9feafd270ef3216a68babdd53c31ea79ae7694 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83180/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5aefcf78ec02e5a24e41b641243ad89167676a1057f2a66b2481774da47669ff +size 45822 diff --git a/ASE_exported/data/atek/cubercnn/train/83325/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83325/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..e95338e343bb4ba00d588ac8a8b6f1f7cb337f09 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83325/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6797baa89a34dd15982cd3b78ebf04b0a784b26aa6ccc3591b55426a74052ff2 +size 31633 diff --git a/ASE_exported/data/atek/cubercnn/train/83346/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83346/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..4465b36f73fdc5cb72e7689b6f581b90e97db734 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83346/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3597feb5c3a1b86eb7a11673c80e6bd85ae9df66b88df98f6e178d9c3e0cdd48 +size 9995 diff --git a/ASE_exported/data/atek/cubercnn/train/83510/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83510/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..1d67c98ba42224559a399ab4a26854a86fb5836e --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83510/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4cbc308412bfbbcf76e8533425103b6d8e712e5cba1131d325c22b13eceedb8 +size 55952 diff --git a/ASE_exported/data/atek/cubercnn/train/83722/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83722/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..2e1be1190a89e7f4fff961c3d82dbc84c7055014 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83722/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc96367e95c92271f37ebd3775dc0009ea8f503f52664d4ca05a1054d5558ff9 +size 19930 diff --git a/ASE_exported/data/atek/cubercnn/train/83948/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83948/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..aae3ef6997bac976018e59c558de6fa70f4fb45a --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83948/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92277d7bc6e24319471910c6ac51109c4d5a51ed70f1ef22883f3172ead8785d +size 8954 diff --git a/ASE_exported/data/atek/cubercnn/train/83977/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/83977/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..236388f4414b399deb83c0f8e1d5894cac57a6fc --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/83977/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15ef4e696b85d0c7265b162955083be73d22a7a0e09f07bcb238361e9bc50f58 +size 22411 diff --git a/ASE_exported/data/atek/cubercnn/train/84100/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84100/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..e918679a4b7adfd81e89a8dce7f69fa9f14ef90f --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84100/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5eec99114651982fc069dc5847a01aedaa6f179fe88dbc4bd552c5420578032 +size 45347 diff --git a/ASE_exported/data/atek/cubercnn/train/84163/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84163/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..104aafccf95035dff21c7c59e5e4784887c309ea --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84163/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95a9c57b1a60b752db5664083c2b763097a166a333bf33b3d1b9f018c7aa4a10 +size 33189 diff --git a/ASE_exported/data/atek/cubercnn/train/84197/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84197/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..0fd00422faad77583740a1a5382c0d5a0f8370e5 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84197/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7afc8019b4ef1bd492652140ccdc259d10ecba4e338f4f5858f2c2b8a07370c +size 63667 diff --git a/ASE_exported/data/atek/cubercnn/train/84332/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84332/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..b9b84d26060e287de356fb19805229673371df95 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84332/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0648f7508c1410155dfb648511d7307261115386f07f4d0d81c8d1f2a7eac4ca +size 22190 diff --git a/ASE_exported/data/atek/cubercnn/train/84351/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84351/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..b36115af57f88494e09e5d6208bde1b0d629ae51 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84351/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91ee5e9902357a5858bc5e0c3732ae08ef7fd3d174e42c617cf441e5d7ff8750 +size 47970 diff --git a/ASE_exported/data/atek/cubercnn/train/84507/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84507/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..eb0a4de2b568d84c2040e9353e9d5947893e734d --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84507/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e899c5297e9f1d2cbd8c3c29d4d67038bf0ebb737611ddb2a316d588c041d6d0 +size 46470 diff --git a/ASE_exported/data/atek/cubercnn/train/84598/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84598/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..30858f066e3ed5f0bb8c1bc77ec3a46f57b205c6 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84598/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21806b2665976ae17929591821c2d7228d453a651f26625ed64b36bf6eaa64bd +size 22015 diff --git a/ASE_exported/data/atek/cubercnn/train/84735/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/84735/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..e267ddb070a25ccbe720e1b3f73cbba87b391cc7 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/84735/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c47786ede0f1da56761b5a30d3884fa9a75f2dddddfe4a22c62a952e93b521cb +size 5728 diff --git a/ASE_exported/data/atek/cubercnn/train/85045/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85045/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..9c84a6db7e078f212b3b3d15161d0cf5adb3f4b7 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85045/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cffb1308e1705cc9b88aadf933af0c7c896829fe1b7f459bc60edaa4c5276f0a +size 28049 diff --git a/ASE_exported/data/atek/cubercnn/train/85277/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85277/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..1a4798aea0a660547111a7108a28a912e9370d67 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85277/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ff58e8503c3b210dda76152699d5304440df7b3fb0df88f1416bf2db3097204 +size 52348 diff --git a/ASE_exported/data/atek/cubercnn/train/85283/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85283/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..53aaa4c1bc097c1f9884feff2001aae5c4014886 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85283/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a0bdec809304ee4e0b824e7e051f3d16f27f945b6fb6d96cd09aaefa9c58bc4 +size 25526 diff --git a/ASE_exported/data/atek/cubercnn/train/85421/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85421/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..0302ffe1de10e24a4fc6afc85309262319edb9aa --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85421/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2cc9177f954e87be1e3c5b64bf5fe7c368ebf11ec1d5424ddfdce1d02ac9da0 +size 60193 diff --git a/ASE_exported/data/atek/cubercnn/train/85613/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85613/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..8e212aef3f8f66a59364309de2547d000093c976 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85613/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86ad55ee508e88a575cd0d53146ad70c439d2ebb3d2581922c2110bc5a8e8b28 +size 14358 diff --git a/ASE_exported/data/atek/cubercnn/train/8574/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/8574/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..7d78c0d952412dfc2287822179ede9e2fde4e2ab --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/8574/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a8943c2ebe3d567b511e3e7d5716b902f063de09fdabf8e54a57d43fb3f20b7 +size 18750 diff --git a/ASE_exported/data/atek/cubercnn/train/8580/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/8580/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..864bde17b1af4bab70220c47e765f0967321dd54 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/8580/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea3344d8efc72a08689b7ac34ffe159b8d164d317bd811e46637634d836d970e +size 33242 diff --git a/ASE_exported/data/atek/cubercnn/train/85812/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85812/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..15d9f369d747aeb7b098727ec94761411febf9d1 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85812/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9152f7afe1960c569ec878464f5be3edb9d7ee514daa5a0729e016b182865889 +size 8088 diff --git a/ASE_exported/data/atek/cubercnn/train/85879/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/85879/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..6af6678ff53bd1f988e3c7d068ccb0774f282177 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/85879/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcb5401aafcc585404e3dc75be0b429971ce447d03169b5ba4a4fa765c343c30 +size 33900 diff --git a/ASE_exported/data/atek/cubercnn/train/86001/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86001/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..73c4e94461b3dead37c9611a70a1256056b2560d --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86001/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46787f448feadaff076c26655fa4f2e3e5a7281e264c4ebd5b96369f26fd1692 +size 21506 diff --git a/ASE_exported/data/atek/cubercnn/train/86062/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86062/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..6af457a29986c5ce8f7926ccd1faf54a85eb4b31 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86062/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a002963a68000a45a15e1d43ded01eed1b789845a167f59b4e8574f5f779510 +size 19318 diff --git a/ASE_exported/data/atek/cubercnn/train/86096/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86096/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..b5993fc7e26c1aa155d2a103a25138f53a6050d5 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86096/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd56a4dc71145021454c969d5a381df9c3c8beed8a1c983ef283b980e7512cca +size 28477 diff --git a/ASE_exported/data/atek/cubercnn/train/86250/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86250/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..0f4239db71f7b24bc13e0488dc48739f7d46c19e --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86250/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:921d1b421e10bc47fc68bdc655db8b7e2eb01b1ff7370f598b3cc6dacc1c2cd2 +size 36737 diff --git a/ASE_exported/data/atek/cubercnn/train/86406/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86406/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..d92fbddeba93950c92f1b9c15b8fe086e145bf2d --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86406/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:798ab12cf61a7e03f7862fd309bad78b6000197b8b80a3bac5f4e4690b67aaf8 +size 21866 diff --git a/ASE_exported/data/atek/cubercnn/train/86439/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86439/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..5b2707b7c91b878aaf5325510748bbab5410c7a7 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86439/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc19835d4298828bafdaf4550ca4eb8f889ded0561d3fa8f466578c5a88c019d +size 31954 diff --git a/ASE_exported/data/atek/cubercnn/train/86465/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86465/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..24fd6f827fcad4d2c04bd97a944b0e6232f6080e --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86465/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94cf56d5b1a3c9dc7e59bb27ecb126c75252d31ef6c9c09c6258f966d050fde9 +size 21161 diff --git a/ASE_exported/data/atek/cubercnn/train/86634/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86634/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..cb3a55f7ffd5796349591f4241cd3a771c560993 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86634/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b80e5d248d047a87b6c4bfb0cf462b7ce16b8acc716cd363d82e747efc2cfc94 +size 31576 diff --git a/ASE_exported/data/atek/cubercnn/train/86657/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/86657/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..95f26ffaadc91aa1f0005088766441e4fb39e764 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/86657/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6c19e30816d3856244fb0c59ad39a2e9f90a6ce9a3df0ebfc30ee27e8e32b9e +size 18806 diff --git a/ASE_exported/data/atek/cubercnn/train/87144/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87144/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..c7d6faaf09207e0f9256dcab3d6fa0ed63660c27 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87144/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bf52e19e42281c0a38466814b85a671087c922c44b74f78e2fdcaabb5a7f4ca +size 4813 diff --git a/ASE_exported/data/atek/cubercnn/train/87376/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87376/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..47a28530aea51827d581388eab8dcba8c6cd8fdb --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87376/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4990a99847f59c90c5cf78e61240345bfcddb7a630e4955b5ed26bb3e9fa90f +size 36695 diff --git a/ASE_exported/data/atek/cubercnn/train/87382/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87382/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..58e33d14f330f1ee21bee8a689783209e1a82d8a --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87382/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a20884ab7f619b58fe64b6392a43a45c5fcd4459ffea9cfbab50b52ce02838ea +size 11553 diff --git a/ASE_exported/data/atek/cubercnn/train/8746/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/8746/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..ac099de870b77f843689534d7926e0d7e20e62a8 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/8746/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b347e7fb36f5838cae8ab6f9e33298c770e197f9865189d428a279b7a898419 +size 23684 diff --git a/ASE_exported/data/atek/cubercnn/train/87520/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87520/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..ed7ddf9c86b603eebce86e92ac53a62c920ead1c --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87520/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cebdfe340a487cfd88b5a3ea94a0a29e6f17191210416a5eabd07c846e4ca40 +size 48039 diff --git a/ASE_exported/data/atek/cubercnn/train/87712/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87712/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..031274bede646e981d80e26dd0a6469a70cf44d6 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87712/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e103f2f305264d462be2dd87c713394d560ed1c41b939af2f5dbea6d830e27a +size 63051 diff --git a/ASE_exported/data/atek/cubercnn/train/87785/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87785/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..f870ca4ebee13526f496e53e4114811be5dc69cb --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87785/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad66c9df60f424b1043cbd096e7903a8674e61312a452a716e0a7c73e18f9f7c +size 7795 diff --git a/ASE_exported/data/atek/cubercnn/train/87978/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/87978/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..131e9f9a8d6df176f9ae19fa4f1c9c40acd4a316 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/87978/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d7e26c866fdd6f1189e853cb2cee6a5102599b532139e4a74981db0517eb0dd +size 54828 diff --git a/ASE_exported/data/atek/cubercnn/train/88355/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/88355/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..ff5f25e5776c8b0e71542a435150003d20aa5702 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/88355/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:677841f305f022f387e3281912e8e81449cf0aedaa11e5b8b42ad58f429d0c1b +size 23774 diff --git a/ASE_exported/data/atek/cubercnn/train/88568/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/88568/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..a03e7f86c81f58c62b1baf7d6a4a9756fded82d7 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/88568/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4108264866e4bb0893e5dfa7662321a9aff2b066aa0091f682a8b3f577fe6cf2 +size 21941 diff --git a/ASE_exported/data/atek/cubercnn/train/88791/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/88791/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..e85440126c6d6953de982ffcfe38feeb61171fe4 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/88791/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:361d5d265516095830d623b76c00df70acddd57a049b17757cab7096e158a13d +size 35605 diff --git a/ASE_exported/data/atek/cubercnn/train/88930/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/88930/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..a76a0ce7244e54f59105b50ebe648ad4ea462608 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/88930/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:930d31b69b63698f27b0f98352c59593ff5172f5aa3c226d7fd5f1eea9bd8b0a +size 21913 diff --git a/ASE_exported/data/atek/cubercnn/train/89218/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/89218/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..d91bee9da3f172c686a49fb9767dd03a281e5192 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/89218/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6da91af7842340c74dcc24d8c43427582d1818b4574951bb460eb82a9c1e7035 +size 6076 diff --git a/ASE_exported/data/atek/cubercnn/train/89287/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/89287/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..2f79a22cd76c81acf4945876294d1d588a0ddec4 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/89287/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f718dd40c97b3fbbea9d163227efa908418df72bc69f3797153ecd10c70b957c +size 13767 diff --git a/ASE_exported/data/atek/cubercnn/train/8947/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/8947/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..80b10c7c81a5c160edb559e4c8b1ed92d7ce9064 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/8947/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c627c57f74b1c913c47df77dc8d32ba2a92ede2cf7ab7910f5941c634920e91 +size 48041 diff --git a/ASE_exported/data/atek/cubercnn/train/89471/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/89471/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..8a02b8cf2554d5f23545d23088840416ad4236e5 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/89471/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eae95b8b68f1d34a7ec985371208b05658becd69c5c24cd82817a8697100d941 +size 8588 diff --git a/ASE_exported/data/atek/cubercnn/train/89643/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/89643/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..ac203702b4e376adee4e042c705e9ed2d61dc311 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/89643/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc2932ca7be16228b670d2320030be1b471cf56b3921eca6606c40ce67e87101 +size 20339 diff --git a/ASE_exported/data/atek/cubercnn/train/89688/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/89688/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..2ff6f305c030d150346703eaa32de8f030e869b3 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/89688/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:176c1bd880a66dad36abc596e3bec8ff7c233aff35938f212057d0a2ea071dd5 +size 23037 diff --git a/ASE_exported/data/atek/cubercnn/train/9036/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/9036/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..c0a11d3a490b022056f8ac92653a085536357eba --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/9036/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db40d25483238a9edc7bce2ec8aa41abb3a23512c80f80f0b075ade702c75885 +size 8519 diff --git a/ASE_exported/data/atek/cubercnn/train/909/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/909/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..29e183a789ef5bc057310570f1feed1782e237fb --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/909/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36a5e1ae628d0a1cb59b3809e1b5692f9cc6046cb605b14a88708d905c015ddf +size 10034 diff --git a/ASE_exported/data/atek/cubercnn/train/9204/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/9204/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..c37522b6a708ba345d43d4a436367e83c8db8e6f --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/9204/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2658443eec4169c7fc5b357b847c4e4306a8bcd026846a35716ea87b3cdc1f4 +size 39741 diff --git a/ASE_exported/data/atek/cubercnn/train/9431/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/9431/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..86258b4c25b7b0b843bab873d19628942670b916 --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/9431/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be5c10390be7f358574b49f61f7d68e23b70385354a3154d2669ea415e94d56e +size 2646 diff --git a/ASE_exported/data/atek/cubercnn/train/9452/annotations.tar.gz b/ASE_exported/data/atek/cubercnn/train/9452/annotations.tar.gz new file mode 100644 index 0000000000000000000000000000000000000000..fc071a102c60c0645c8de7384ffdc5b7fcce680c --- /dev/null +++ b/ASE_exported/data/atek/cubercnn/train/9452/annotations.tar.gz @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d349211255348fde394f4efe73573944d4b6d19f4e31f301b4bba0eb41c4c05 +size 30809 diff --git a/HM3DSem_exported/README.md b/HM3DSem_exported/README.md new file mode 100644 index 0000000000000000000000000000000000000000..91975609143624eff25a76fdd336264c48f1c1cb --- /dev/null +++ b/HM3DSem_exported/README.md @@ -0,0 +1,101 @@ +# HM3DSem 导出 + +## 目录 + +```text +HM3DSem_exported/ +├── README.md +├── data/ +│ ├── hm3d-train-semantic-annots-v0.2.tar +│ ├── hm3d-val-semantic-annots-v0.2.tar +│ ├── hm3d-train-semantic-configs-v0.2.tar +│ ├── hm3d-val-semantic-configs-v0.2.tar +│ └── metadata.csv +├── kits/ +│ ├── convert_hm3dsem.py +│ ├── unify_hm3dsem.py +│ └── verify_hm3dsem.py +├── hm3dsem.jsonl +└── multiroom/json_multi_merged.json +``` + +`data/` 保留完整源标注 TAR、语义配置 TAR 和场景元数据。`kits/` 保留转换、汇总和验收脚本，转换直接读取本目录中的 TAR。 + +`hm3dsem.jsonl` 每行是一个完整场景；`multiroom/json_multi_merged.json` 将相同记录保存为 JSON 数组，不重新计算字段或拆房间。当前共 181 个场景、2,599 个源语义区域、92,425 条家具实例和 27,867 条结构实例；8 个无家具的区域仍保留。 + +本地 `.git/`、`.gitignore` 可保留，上传 Hugging Face 时排除。 + +## 上游源数据 + +源数据为官方 [HM3DSem v0.2](https://aihabitat.org/datasets/hm3d-semantics/) 语义标注，场景元数据来自 [HM3D](https://aihabitat.org/datasets/hm3d/)。本导出使用 train 的 145 个场景和 val 的 36 个场景，不包含 test。 + +每个源场景包含 `semantic.txt` 与 `semantic.glb`：TXT 提供实例 ID、RGB 语义颜色、类别和 region ID；GLB 提供带语义颜色的扫描网格；`metadata.csv` 提供划分及场景属性。两个标注 TAR 共含 362 个场景文件，转换无需外部下载目录。 + +家具几何采用同色网格连通分量中 AABB 体积最大者，位置为该框中心。这一选择参考固定 Habitat-Sim 版本 `57ee4941dc4765240f0f91f70b2c97a919bf9038` 的[分量与包围盒计算](https://github.com/facebookresearch/habitat-sim/blob/57ee4941dc4765240f0f91f70b2c97a919bf9038/src/esp/scene/SemanticScene.cpp#L276-L299)及[最大体积分量选择](https://github.com/facebookresearch/habitat-sim/blob/57ee4941dc4765240f0f91f70b2c97a919bf9038/src/esp/scene/SemanticScene.cpp#L444-L511)；本导出保留原生 Z-up 坐标，未声明与 Habitat 运行时逐数值一致。 + +## schema 字段表 + +字段以[转换脚本](kits/convert_hm3dsem.py)为准。长度为米，面积为平方米；坐标为右手 Z-up、场景全局原点、原生 semantic GLB 坐标，不将家具平移到地板。字段缺失时保留键并写 `null`。 + +### 场景字段 + +| 字段 | 类型 | 来源与取值 | +|---|---|---| +| `scene_id` | string | 原始场景目录名，保留数字前缀，如 `<序号>-`。 | +| `sample_id` | string | 与 `scene_id` 相同。 | +| `dataset` | string | `HM3DSem`。 | +| `source_dataset` | string | `HM3D`。 | +| `source_subset` | string / null | 仅取 `metadata.csv.split`，当前为 `train` 或 `val`；缺失则 `null`。 | +| `room_count` | integer | `len(rooms)`，不是元数据的 `num_rooms`。 | +| `total_area` | number / null | 累加已知房间面积，全部缺失则 `null`；部分缺失时只是已知部分之和。 | +| `building_style` | null | 不生成风格标签。 | +| `description` | null | 不生成描述。 | +| `description_source` | null | 无描述来源。 | +| `provenance` | object | `source_file` 为相对导出根目录的 `data/<标注TAR>::<场景>/.semantic.txt`；`source_value` 保留元数据中除已映射 `scene`、`split` 外的字段原字符串，无元数据行为 `null`。 | +| `rooms` | array | 按原始 region ID 数值升序分组，包含无家具区域。 | + +### 房间字段 + +| 字段 | 类型 | 来源与取值 | +|---|---|---| +| `room_id` | string | `::region_<原始region ID>`。 | +| `room_index` | integer | region 排序后的连续索引，从 0 开始。 | +| `room_type` | string / null | 项目决定：从同一区域的源类别按脚本中的精确词表、组合和优先级推断；19 类及 `null`，见下文。 | +| `room_type_source` | string | 有推断为 `inferred_from_furniture`，否则 `unavailable`。 | +| `room_boundary` | array<[number, number]> / null | 区域内精确类别 `floor` 的可用网格顶点 XY 凸包；无有效、无歧义的凸包则 `null`。 | +| `room_boundary_type` | string / null | 有边界为 `floor_mesh_convex_hull`，否则 `null`。 | +| `room_area_m2` | number / null | 同一凸包的鞋带面积；不使用元数据的 `floor_space` 或 `navigable_area` 补值。 | +| `room_height` | number / null | 精确类别 `ceiling` 的最大 Z 减 `floor` 的最小 Z；源缺失、有歧义、非有限或非正值时为 `null`。 | +| `coordinate_system` | string | 固定说明米制、右手 Z-up、原生 GLB 全局坐标、最大体积分量 AABB 中心及世界轴尺寸、方向未知。 | +| `doors` | array | `{furniture_index:n}`，指向本房间家具数组中精确类别为 `door` 的条目；无记录为 `[]`。 | +| `windows` | array | 同 `doors`，精确类别为 `window`。 | +| `furniture_quantity` | integer | `len(furniture)`，含几何缺失的实例。 | +| `furniture` | array | 按实例 ID 排序，排除下述六个精确结构类别。 | +| `source_fields` | object | `structural_instances` 按原类别名分组，每组保存 `{id,color}` 数组；无结构实例时为 `{"structural_instances":{}}`。 | + +结构类别为 `wall`、`floor`、`ceiling`、`shower wall`、`shower floor`、`shower ceiling`。房型推断读取分组前的同区域源类别，比较时去除首尾空白并忽略大小写；结构分类、地板/天花几何与门窗引用仍按各自的精确类别规则处理。 + +房型取值为 `bedroom`、`living room`、`dining room`、`kitchen`、`bathroom`、`office`、`laundry room`、`storage room`、`recreation room`、`gym`、`spa`、`garage`、`hallway`、`stairwell`、`outdoor`、`living dining room`、`kitchen dining room`、`living kitchen room`、`living kitchen dining room`，无法判断为 `null`。这些不是源数据给出的房型标签。 + +### 家具字段 + +| 字段 | 类型 | 来源与取值 | +|---|---|---| +| `furniture_category` | string | 原类别字符串，不改写拼写，不做跨数据集归类，保留源 `unknown`。 | +| `furniture_instance_id` | string | `:<原始实例ID>`，此处为单冒号。 | +| `furniture_position` | object / null | 所选世界轴 AABB 的中心 `{x,y,z}`，每轴为 `(min+max)/2`；无可用、无歧义实例框则 `null`。 | +| `furniture_rotation` | object | 始终为 `{x:null,y:null,z:null}`，AABB 不提供物体朝向。 | +| `furniture_size` | object / null | 同一 AABB 的完整 X/Y/Z 跨度对应 `{width,length,height}`；与位置同时缺失。 | +| `source_fields` | object | 仅 `color`，保留源 1–6 位 RGB 十六进制字符串，不补零。 | + +同色对应多个源实例时不擅自分配实例框；最大体积并列且框极值不同也记为空。并列但框相同可保留该框。无顶点时位置和尺寸均为空；零跨度尺寸原样保留。地板/天花共享颜色只有在类别与区域归属一致、无歧义时才用于房间几何。 + +`provenance.source_value` 保留 `furnished`、`navmesh_quality`、`large_holes`、`portals_blocked`、`low_interactive_potential`、`doors_scanned_closed`、`unfurnished`、`num_floors`、`overall_scene_quality`、`diversity`、`num_rooms`、`navigable_area`、`navigation_complexity`、`scene_clutter`、`floor_space`。这些源指标与导出房间数量、面积的语义不同。 + +## 已知缺陷 + +- 源 region 是语义区域，不保证等于完整建筑房间。扫描不完整，地板凸包可能填平凹角或覆盖未扫描区域；层高是扫描垂直跨度，不是已确认的净高。 +- 家具 AABB 描述所选扫描分量的世界轴范围，不是物体局部尺寸、质心或完整姿态。当前 2,431 条家具位置和尺寸为空：2,411 条缺少顶点，19 条颜色重复，1 条最大体积分量有歧义；另有 3 条零跨度框。 +- 当前 188 个区域边界和面积为空、260 个层高为空；`total_area` 可能只覆盖部分区域，不能当作源场景真实总面积。 +- 房型为规则推断，尚无独立标注集准确率。工作桌与衣物储藏的优先级、衣物数量达到工作桌数量 10 倍等条件是项目启发式；当前 521 个区域房型为空。 +- 门窗只引用精确类别 `door`、`window`，不自动纳入其他拼写或复合类别。家具与结构实例保留不代表扫描覆盖完整实物。 diff --git a/HM3DSem_exported/kits/convert_hm3dsem.py b/HM3DSem_exported/kits/convert_hm3dsem.py new file mode 100644 index 0000000000000000000000000000000000000000..bad2a78c31318ea93b40959bcc45dc1bcc53816f --- /dev/null +++ b/HM3DSem_exported/kits/convert_hm3dsem.py @@ -0,0 +1,453 @@ +#!/usr/bin/env python3 +"""Convert retained HM3DSem v0.2 archives to one complete scene per JSONL line. + +Use --limit or --scene-id for smoke runs, with --out in a temporary directory. +Full conversion requires --all and prior user approval. Diagnostics go to stdout. +""" +from __future__ import annotations + +import argparse +from collections import Counter, defaultdict +from contextlib import ExitStack +import csv +import io +import json +import math +import os +from pathlib import Path +import struct +import sys +import tarfile +import time + +import numpy as np +from scipy.sparse import coo_matrix +from scipy.sparse.csgraph import connected_components + +BASE = Path(__file__).resolve().parents[1] +STRUCTURAL = frozenset({'wall', 'floor', 'ceiling', 'shower wall', 'shower floor', 'shower ceiling'}) +COORDINATES = ('meter; right-handed; Z-up; scene-global origin; native semantic GLB coordinates; ' + 'position = largest-volume semantic connected-component AABB center; ' + 'size = world-axis AABB extents; object orientation unavailable') +METADATA_KEYS = ('furnished navmesh_quality large_holes portals_blocked low_interactive_potential ' + 'doors_scanned_closed unfurnished num_floors overall_scene_quality diversity ' + 'num_rooms navigable_area navigation_complexity scene_clutter floor_space').split() + +# User-approved 19 room types + null, 2026-09-21. Exact same-region categories. +GROUPS = { + 'BED': {'bed', 'bed small', 'bunk bed', 'bedframe', 'bed base'}, + 'BEDDING': {'bed comforter', 'bed sheet', 'bedding', 'duvet'}, + 'BEDSIDE': {'nightstand', 'bedside cabinet', 'bedside table', 'bed table', 'bed stand'}, + 'BATH_FIXTURE': {'bath', 'bath sink', 'bath tub', 'bathtub', 'bidet', 'shower', 'shower cabin', 'shower cockpit', 'shower stall', 'shower tub', 'toilet', 'urinal', 'washbasin'}, + 'SHOWER_WATER': {'showerhead', 'shower hose/head', 'shower hose', 'shower tap', 'shower battery', 'shower valve', 'shower dial', 'shower knob', 'shower pipe', 'shower handle'}, + 'SHOWER_SURFACE': {'shower wall', 'shower floor', 'shower tray', 'shower base', 'shower glass', 'shower door', 'shower curtain', 'shower ceiling', 'shower seat', 'shower bench', 'shower frame', 'shower door frame', 'shower step', 'shower bar'}, + 'BATH_VANITY': {'bath cabinet', 'bathroom cabinet', 'bathroom counter', 'bathroom shelf', 'wash cabinet', 'washbasin counter', 'bath shelf', 'bathroom cabinet door', 'bathroom cabinet drawer', 'shower-bath cabinet'}, + 'WATER': {'sink', 'sink/basin', 'basin', 'tap', 'faucet'}, + 'BASIN': {'sink', 'sink/basin', 'basin'}, + 'WASHER': {'clothes dryer', 'laundry machine', 'washer-dryer', 'washing machine', 'washing machine and dryer'}, + 'WORK_EQUIPMENT': {'typewriter', 'computer equipment', 'monitor', 'computer tower', 'laptop', 'copier machine', 'printer', 'copier', 'pc tower', 'computer', 'scanner'}, + 'WORK_SURFACE': {'desk', 'computer desk', 'office table', 'workstation'}, + 'WORK_SUPPORT': {'desk chair', 'office chair', 'computer chair', 'computer', 'laptop', 'monitor', 'computer tower', 'pc tower', 'printer', 'copier', 'copier machine', 'scanner', 'file cabinet', 'typewriter', 'desk clutter', 'stationery', 'computer equipment', 'computer mouse', 'mouse', 'mousepad', 'document', 'document holder', 'folder', 'binder', 'binders', 'file binder', 'stack of binders', 'desk organizer', 'pencil holder'}, + 'BOOK_STORAGE': {'bookshelf', 'book rack', 'book cabinet'}, + 'BOOKS': {'book', 'stack of books'}, + 'CLOTHING_FACILITY': {'closet', 'closet area for hanging clothes', 'closet storage area', 'closet floor', 'closet rod', 'closet shelf', 'closet shelving', 'wardrobe', 'clothes rack', 'clothes hanger rod', 'drawers for clothes', 'clothes container', 'clothing stand', 'rack', 'shelf', 'shelving', 'clothes hanger', 'cloth hanger', 'cloth hangers', 'coat hanger', 'hanger', 'set of hangers', 'coat rack', 'shoe rack', 'shoes rack', 'shoe shelf', 'rack with shoes', 'shelf with shoes'}, + 'CLOTHING_CONTENTS': {'clothes', 'hanging clothes', 'stack of clothes', 'clothes on shelf', 'sheets / clothes', 'shirt', 't-shirt', 'stack of t-shirts', 'coat', 'jacket', 'stack of jackets', 'dress', 'blouse', 'socks', 'shoe', 'shoes', 'shoes on shelf', 'stack of shoes'}, + 'EXPLICIT_STORAGE': {'storage', 'storage space', 'storage unit', 'storage shelving', 'storage cabinet', 'closet storage area', 'pantry', 'wardrobe', 'closet', 'closet shelf', 'closet shelving', 'closet rod', 'clothes hanger rod'}, + 'CLOSET_AREA': {'closet area for hanging clothes', 'closet storage area', 'closet floor'}, + 'HALLWAY_ANCHOR': {'ceiling corridor'}, + 'STAIR_ANCHOR': {'stairs', 'stair', 'staircase', 'stairwell', 'stair step', 'bottom of stairs', 'landing'}, + 'OUTDOOR_ANCHOR': {'balcony', 'patio', 'patio floor', 'garden deck', 'terrace', 'floor /outside'}, + 'K_DIRECT_COOKING': {'cooker', 'oven', 'oven and stove', 'stove', 'stovetop'}, + 'K_FIXED_INFRASTRUCTURE': {'cabinet kitchen', 'cooker hood', 'extractor hood', 'kitchen cabinet', 'kitchen cabinet door', 'kitchen cabinet drawer', 'kitchen cabinet lower', 'kitchen counter', 'kitchen counter support', 'kitchen extractor', 'kitchen island', 'kitchen lower cabinet', 'kitchen sink', 'kitchen sink cabinet', 'kitchen top', 'range hood', 'ventilation hood', 'worktop'}, + 'L_DIRECT_SOFA': {'circular sofa', 'couch', 'l-shaped sofa', 'sofa', 'sofa chair', 'sofa set'}, + 'L_SOFA_PART': {'sofa seat'}, + 'L_SUPPORT': {'coffee table', 'led tv', 'pouffe', 'tv', 'tv stand', 'wall tv'}, + 'D_EXPLICIT_TABLE': {'dining table', 'dinner table'}, + 'D_GENERIC_TABLE': {'table', 'kitchen table'}, + 'D_SEAT': {'bar chair', 'chair', 'dining chair', 'dinner chair', 'kitchen chair', 'stool', 'stools'}, + 'GYM_MACHINE': {'treadmill', 'exercise bike', 'workout bike', 'cross-trainer', 'gym stepper', 'exercise machine', 'weight bench', 'boxing ring', 'punchbag', 'rack of weights'}, + 'GYM_GENERAL': {'gym equipment', 'exercise equipment'}, + 'GYM_SUPPORT': {'barbell', 'dumbbell', 'weights', 'gym mat', 'exercise mat', 'training mat', 'yoga mat', 'gym rope'}, + 'SPA_DIRECT': {'sauna', 'sauna heater', 'sauna oven', 'sauna heat rocks', 'foot spa', 'spa bathtub', 'jacuzzi', 'massage bed'}, + 'SPA_SEAT': {'spa bench', 'spa armchair'}, + 'SPA_WATER': {'washbasin', 'showerhead', 'shower hose', 'shower hose/head', 'spa bathtub', 'foot spa'}, + 'GAME': {'air hockey', 'arcade', 'arcade game', 'foosball game table', 'foosball table', 'ping pong table', 'pool table', 'table tennis table', 'ball pool'}, + 'SCREEN': {'panel screen', 'projector screen', 'screen'}, + 'MEDIA_SEAT': {'armchair', 'row of theater chairs', 'couch', 'sofa'}, + 'GARAGE_MOTOR': {'garage door motor', 'garage door opener', 'garage door opener motor'}, + 'VEHICLE': {'car', 'motorcycle'}, + 'BARBECUE': {'grill', 'barbecue'}, + 'OUTDOOR_CONTEXT': {'patio chair', 'roof'}, + 'PLAY_FURNITURE': {'table', 'chair', 'pouffe'}, + 'STORAGE_AREA': {'storage', 'storage space', 'pantry', 'closet storage area'}, + 'BATHING': {'bath', 'bath tub', 'bathtub', 'shower', 'shower cabin', 'shower cockpit', 'shower stall', 'shower tub'}, +} + +def room_signals(c): + h=lambda key: not GROUPS[key].isdisjoint(c) + office = h('WORK_SURFACE') or (h('D_GENERIC_TABLE') and h('WORK_SUPPORT')) or (h('BOOK_STORAGE') and h('BOOKS')) + f={ + 'B':h('BED') or h('BEDDING') and h('BEDSIDE'), + 'BATH':h('BATH_FIXTURE') or h('SHOWER_SURFACE') or h('BATH_VANITY') and h('WATER'), + 'BATH_WEAK':h('BASIN') and 'mirror' in c, + 'K':h('K_DIRECT_COOKING') or h('K_FIXED_INFRASTRUCTURE'), + 'L':h('L_DIRECT_SOFA') or h('L_SOFA_PART') and h('L_SUPPORT') or c.get('armchair',0)>=2 and bool({'table','coffee table'}&c.keys()), + 'O':office, + 'W':h('WASHER'), + 'S':h('EXPLICIT_STORAGE') or h('CLOSET_AREA') or h('CLOTHING_FACILITY') and h('CLOTHING_CONTENTS'), + 'T':h('STAIR_ANCHOR'),'H':h('HALLWAY_ANCHOR'), + 'X':h('OUTDOOR_ANCHOR') or h('BARBECUE') and h('OUTDOOR_CONTEXT') and 'garage door' not in c, + 'G':h('GYM_MACHINE') or h('GYM_GENERAL') and h('GYM_SUPPORT'), + 'P':h('SPA_DIRECT') or h('SPA_SEAT') and h('SPA_WATER'), + 'R':h('GAME') or 'projector' in c and h('SCREEN') and 'speaker' in c and h('MEDIA_SEAT'), + 'A':'garage door' in c and (h('GARAGE_MOTOR') or h('VEHICLE')), + } + # Project heuristic: substantial clothing storage can outweigh work desks. + desks = sum(c.get(x, 0) for x in GROUPS['WORK_SURFACE']) + if desks and h('CLOTHING_FACILITY') and sum(c.get(x, 0) for x in GROUPS['CLOTHING_CONTENTS']) >= 10 * desks and not h('WORK_EQUIPMENT'): + f['O'] = False + if (f['T'] or f['H']) and not (h('WORK_SUPPORT') or h('BOOK_STORAGE') and h('BOOKS')): + f['O'] = False + # Work-area tables/chairs do not establish dining without kitchen evidence. + f['D'] = h('D_EXPLICIT_TABLE') or (h('D_GENERIC_TABLE') and sum(c.get(x,0) for x in GROUPS['D_SEAT']) >= 4 and not (office and not f['K'] and (not f['L'] or h('WORK_SURFACE')))) + # Toys plus activity furniture can outrank a storage cabinet, but not a + # separately evidenced bedroom, work area, circulation or clothes store. + f['R']=f['R'] or ( + c.get('toy',0)>=2 and h('PLAY_FURNITURE') + and not any(f[k] for k in ('B','BATH','BATH_WEAK','K','L','D','O','W','T','H','X','G','P','A')) + and not h('STORAGE_AREA') and not h('CLOSET_AREA') + and not (h('CLOTHING_FACILITY') and h('CLOTHING_CONTENTS')) + ) + return f +ROOM_TYPES = {'B':'bedroom','BATH':'bathroom','O':'office','W':'laundry room','S':'storage room','T':'stairwell','H':'hallway','X':'outdoor','G':'gym','P':'spa','R':'recreation room','A':'garage'} +SOCIAL_TYPES = {frozenset(k):v for k,v in [('K','kitchen'),('L','living room'),('D','dining room'),('KL','living kitchen room'),('KD','kitchen dining room'),('LD','living dining room'),('KLD','living kitchen dining room')]} +BATHING = set(GROUPS['BATHING']) +def room_type(rows): + c = Counter(r['category'].strip().casefold() for r in rows) + f=room_signals(c);a={k for k,v in f.items() if v and k!='BATH_WEAK'} + if 'X' in a:a-=set('KLD') + if 'B' in a and 'garage door' not in c:a-=set('LOSGT') + if 'P' in a:a-={'BATH','O','S','T','H'} + if 'R' in a:a.discard('L') + if 'K' in a:a.discard('W') + if 'BATH' in a and (bool(BATHING & c.keys()) or bool(GROUPS['SHOWER_WATER']&c.keys()) and bool(GROUPS['SHOWER_SURFACE']&c.keys())):a.discard('W') + if a&set('KLD'):a-=set('OS') + # Dedicated facilities outrank work furniture; circulation outranks storage. + if a & {'BATH','G','R','W'}: + a.discard('O') + if a & {'T','H'}: + a.discard('S') + if a-{'S','T','H'}: + a.discard('S') + # Washer presence cannot establish whether circulation is ancillary. + if 'W' not in a:a-={'T','H'} + if frozenset(a) in SOCIAL_TYPES:out=SOCIAL_TYPES[frozenset(a)] + elif len(a)==1:out=ROOM_TYPES[next(iter(a))] + elif not a and f['BATH_WEAK']:out='bathroom' + else:out=None + return out + + +def source_index(data, stack): + """Seekable TARs stay open; only selected member payloads are read.""" + result = {} + for split in ('train', 'val'): + archive_name = f'hm3d-{split}-semantic-annots-v0.2.tar' + archive = stack.enter_context(tarfile.open(data / archive_name)) + pairs = defaultdict(dict) + for member in archive: + if not member.isfile(): + continue + sid, filename = member.name.split('/') + pairs[sid][filename.rsplit('.', 1)[1]] = member + for sid, members in pairs.items(): + if sid in result: + raise ValueError(f'duplicate scene: {sid}') + result[sid] = (archive, archive_name, members) + return result + + +def read_metadata(path): + with path.open(newline='', encoding='utf-8') as handle: + reader = csv.DictReader(handle) + return {row['scene']: row for row in reader} + + +def annotation_rows(raw): + header, sep, body = raw.decode('utf-8').partition('\n') + if header.rstrip('\r') != 'HM3D Semantic Annotations' or not sep: + raise ValueError('invalid semantic TXT header') + rows, seen = [], set() + for cells in csv.reader(io.StringIO(body), strict=True): + if not cells: + continue + ident, color, category, region = cells + ident, region = int(ident), int(region) + int(color, 16) # Official parser accepts hexadecimal short forms. + if ident < 0 or region < 0 or ident in seen or not category.strip(): + raise ValueError(f'invalid or duplicate annotation: {cells}') + seen.add(ident) + rows.append(dict(id=ident, color=color, category=category, region=region)) + return sorted(rows, key=lambda row: row['id']) + + +def read_glb(raw): + """Decode the observed indexed FLOAT VEC3 / normalized USHORT VEC4 release. + + This is a reader for the inspected release, not a general glTF validator. + Non-mesh leaf nodes (the release includes a light and a camera) are inert. + """ + if len(raw) < 28 or struct.unpack_from('<4sII', raw) != (b'glTF', 2, len(raw)): + raise ValueError('invalid GLB header/length') + n, kind = struct.unpack_from(' len(raw): + raise ValueError('invalid GLB JSON chunk') + doc = json.loads(raw[20:20+n]) + bn, bk = struct.unpack_from(' len(blob) or + needed > length): + raise ValueError('accessor outside buffer bounds') + return np.ndarray((count, width), dtype=dtype, buffer=blob, + offset=start+offset, strides=(stride, dtype.itemsize)) + + nodes = [doc['nodes'][i] for i in doc['scenes'][doc.get('scene', 0)]['nodes']] + meshes = [] + for node in nodes: + if node.get('children') or ('mesh' in node and (node.get('extensions') or set(node) & {'matrix', 'rotation', 'translation', 'scale', 'skin', 'weights'})): + raise ValueError('unsupported mesh transform/hierarchy') + if 'mesh' in node: + meshes.append(node['mesh']) + vertices, colors, faces, offset = [], [], [], 0 + for mesh_index in meshes: + for primitive in doc['meshes'][mesh_index]['primitives']: + if primitive.get('mode', 4) != 4 or primitive.get('targets') or primitive.get('extensions'): + raise ValueError('unsupported primitive') + attrs = primitive['attributes'] + v = accessor(attrs['POSITION'], 'VEC3', {5126}) + c = accessor(attrs['COLOR_0'], 'VEC4', {5123}, True) + f = accessor(primitive['indices'], 'SCALAR', {5121, 5123, 5125}).reshape(-1) + if len(c) != len(v) or len(f) % 3 or (len(f) and int(f.max()) >= len(v)): + raise ValueError('invalid vertex/index counts') + vertices.append(v); colors.append(c); faces.append(f.reshape(-1, 3).astype(np.int64)+offset) + offset += len(v) + if not vertices: + return np.empty((0, 3)), np.empty((0, 4), dtype=np.uint16), np.empty((0, 3), dtype=np.int64) + return np.concatenate(vertices).astype(np.float64), np.concatenate(colors), np.concatenate(faces) + + +def color_values(rawcolors): + linear = rawcolors[:, :3].astype(np.float64) / 65535 + srgb = np.where(linear <= 0.0031308, linear*12.92, 1.055*linear**(1/2.4)-0.055) + rgb = np.floor(srgb*255+0.5).clip(0, 255).astype(np.uint32) + return (rgb[:, 0] << 16) | (rgb[:, 1] << 8) | rgb[:, 2] + + +def geometry(vertices, rawcolors, faces, rows, diagnostics): + colors = color_values(rawcolors) + invalid_colors = set(map(int, colors[~np.isfinite(vertices).all(axis=1)])) + if invalid_colors - {int(r['color'], 16) for r in rows}: + raise ValueError('nonfinite geometry cannot be localized to source annotations') + # Poison the whole affected color; never build a box from its remaining points. + good = ~np.isin(colors, list(invalid_colors)) + old_to_good = np.full(len(vertices), -1, dtype=np.int64) + old_to_good[good] = np.arange(int(good.sum())) + # Retain valid-valid edges even when the third triangle vertex is invalid. + edges = np.concatenate((faces[:, [0, 1]], faces[:, [1, 2]], faces[:, [2, 0]])) + edges = edges[good[edges].all(axis=1)] + edges = old_to_good[edges] + v, c = vertices[good], rawcolors[good] + if not len(v): + return {}, {}, invalid_colors + # Exact position + original color welding precedes linear-to-sRGB conversion. + key = np.column_stack((v, c.astype(np.float64))) + _, first, inverse = np.unique(key, axis=0, return_index=True, return_inverse=True) + v, colors = v[first], colors[good][first] + edges = inverse[edges] + edges = edges[colors[edges[:, 0]] == colors[edges[:, 1]]] + graph = coo_matrix((np.ones(len(edges), dtype=bool), (edges[:, 0], edges[:, 1])), + shape=(len(v), len(v))).tocsr() + count, labels = connected_components(graph, directed=False) + low, high = np.full((count, 3), np.inf), np.full((count, 3), -np.inf) + np.minimum.at(low, labels, v); np.maximum.at(high, labels, v) + component_colors = np.zeros(count, dtype=np.uint32) + component_colors[labels] = colors + boxes, points = {}, {} + for color in np.unique(colors): + color = int(color) + points[color] = v[colors == color] + ids = np.flatnonzero(component_colors == color) + sizes = high[ids]-low[ids] + volumes = np.prod(sizes, axis=1) + if not np.isfinite(volumes).all() or (sizes < 0).any(): + raise ValueError('invalid component extrema/volume') + winners = ids[volumes == volumes.max()] + best = winners[0] + if not (np.all(low[winners] == low[best]) and np.all(high[winners] == high[best])): + diagnostics['ambiguous_component_colors'].append(f'{color:06X}') + boxes[color] = None + else: + boxes[color] = (low[best], high[best]) + return boxes, points, invalid_colors + + +def convex_hull(points): + """Monotone chain, using unrounded coordinates and exact turn comparisons.""" + points = sorted(set(map(tuple, points))) + if len(points) < 3: + return None + def cross(a, b, c): + return (b[0]-a[0])*(c[1]-a[1]) - (b[1]-a[1])*(c[0]-a[0]) + def chain(seq): + out = [] + for p in seq: + while len(out) >= 2 and cross(out[-2], out[-1], p) <= 0: + out.pop() + out.append(p) + return out + hull = chain(points)[:-1] + chain(reversed(points))[:-1] + return [list(p) for p in hull] if len(hull) >= 3 else None + + +def room_geometry(rows, by_color, points, invalid_colors): + def surface(category): + selected = {int(r['color'], 16) for r in rows if r['category'] == category} + bad = any(c in invalid_colors or len({(r['region'], r['category']) for r in by_color[c]}) > 1 + for c in selected) + values = [points[c] for c in sorted(selected) if c in points] + return (np.concatenate(values) if values else np.empty((0, 3))), bad + floors, bad_floor = surface('floor') + ceilings, bad_ceiling = surface('ceiling') + boundary = convex_hull(floors[:, :2]) if not bad_floor else None + area = None + if boundary is not None: + area = abs(sum(a[0]*b[1]-a[1]*b[0] for a, b in zip(boundary, boundary[1:]+boundary[:1])))/2 + if not math.isfinite(area) or area <= 0: + boundary, area = None, None + height = None + if len(floors) and len(ceilings) and not (bad_floor or bad_ceiling): + value = float(ceilings[:, 2].max()-floors[:, 2].min()) + if math.isfinite(value) and value > 0: + height = value + return boundary, area, height + + +def convert_scene(sid, rows, mesh, metadata, locator): + diagnostics = {'ambiguous_component_colors': [], 'duplicate_colors': [], + 'nonfinite_colors': [], 'missing_geometry': [], 'degenerate_instances': []} + by_color, regions = defaultdict(list), defaultdict(list) + for row in rows: + by_color[int(row['color'], 16)].append(row) + regions[row['region']].append(row) + diagnostics['duplicate_colors'] = [f'{c:06X}' for c, rs in sorted(by_color.items()) if len(rs) > 1] + boxes, points, invalid = geometry(*mesh, rows, diagnostics) + diagnostics['nonfinite_colors'] = [f'{c:06X}' for c in sorted(invalid)] + rooms = [] + for index, (region, rr) in enumerate(sorted(regions.items())): + furniture, structural = [], defaultdict(list) + for row in rr: + if row['category'] in STRUCTURAL: + structural[row['category']].append({'id': row['id'], 'color': row['color']}) + continue + color = int(row['color'], 16) + box = boxes.get(color) if len(by_color[color]) == 1 else None + position = size = None + if box is not None: + low, high = box + position = dict(zip(('x', 'y', 'z'), ((low+high)/2).tolist())) + size = dict(zip(('width', 'length', 'height'), (high-low).tolist())) + if np.any(high == low): + diagnostics['degenerate_instances'].append(row['id']) + elif color not in points and color not in invalid: + diagnostics['missing_geometry'].append(row['id']) + furniture.append({'furniture_category': row['category'], 'furniture_instance_id': f"{sid}:{row['id']}", + 'furniture_position': position, 'furniture_rotation': dict.fromkeys(('x', 'y', 'z')), + 'furniture_size': size, 'source_fields': {'color': row['color']}}) + boundary, area, height = room_geometry(rr, by_color, points, invalid) + prediction = room_type(rr) + rooms.append({'room_id': f'{sid}::region_{region}', 'room_index': index, + 'room_type': prediction, 'room_type_source': 'inferred_from_furniture' if prediction else 'unavailable', + 'room_boundary': boundary, 'room_boundary_type': 'floor_mesh_convex_hull' if boundary is not None else None, + 'room_area_m2': area, 'room_height': height, 'coordinate_system': COORDINATES, + 'doors': [{'furniture_index': i} for i, f in enumerate(furniture) if f['furniture_category'] == 'door'], + 'windows': [{'furniture_index': i} for i, f in enumerate(furniture) if f['furniture_category'] == 'window'], + 'furniture_quantity': len(furniture), 'furniture': furniture, + 'source_fields': {'structural_instances': dict(sorted(structural.items()))}}) + known = [r['room_area_m2'] for r in rooms if r['room_area_m2'] is not None] + split = (metadata['split'] or None) if metadata is not None else None + if split not in {'train', 'val', None}: + raise ValueError('selected source scene has unsupported metadata split') + scene = {'scene_id': sid, 'sample_id': sid, 'dataset': 'HM3DSem', 'source_dataset': 'HM3D', + 'source_subset': split, 'room_count': len(rooms), 'total_area': sum(known) if known else None, + 'building_style': None, 'description': None, 'description_source': None, + 'provenance': {'source_file': locator, + 'source_value': {k: metadata[k] for k in METADATA_KEYS} if metadata is not None else None}, + 'rooms': rooms} + return scene, diagnostics + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('--data-dir', type=Path, default=BASE/'data') + parser.add_argument('--out', type=Path, default=BASE/'hm3dsem.jsonl') + scope = parser.add_mutually_exclusive_group(required=True) + scope.add_argument('--limit', type=int) + scope.add_argument('--scene-id', action='append') + scope.add_argument('--all', action='store_true') + args = parser.parse_args() + if args.limit is not None and args.limit < 1: + parser.error('--limit must be positive') + temp = args.out.with_suffix(args.out.suffix+'.part') + started = time.perf_counter() + report = {'scenes': []} + args.out.parent.mkdir(parents=True, exist_ok=True) + with ExitStack() as stack: + index = source_index(args.data_dir, stack) + metadata = read_metadata(args.data_dir/'metadata.csv') + selected = sorted(set(args.scene_id)) if args.scene_id else sorted(index)[:args.limit] + if not selected or set(selected)-set(index): + raise ValueError('empty selection or unknown scene ID') + with temp.open('w', encoding='utf-8') as output: + for current in selected: + begin = time.perf_counter() + archive, archive_name, members = index[current] + rows = annotation_rows(archive.extractfile(members['txt']).read()) + mesh = read_glb(archive.extractfile(members['glb']).read()) + scene, diagnostics = convert_scene(current, rows, mesh, metadata.get(current), + f"data/{archive_name}::{members['txt'].name}") + output.write(json.dumps(scene, ensure_ascii=False, allow_nan=False)+'\n') + report['scenes'].append({'scene_id': current, 'seconds': time.perf_counter()-begin, + 'glb_bytes': members['glb'].size, 'vertices': len(mesh[0]), + 'source_instances': len(rows), 'rooms': len(scene['rooms']), + 'furniture': sum(r['furniture_quantity'] for r in scene['rooms']), + 'diagnostics': diagnostics}) + print(f'{current}: converted in {report["scenes"][-1]["seconds"]:.3f}s', file=sys.stderr, flush=True) + del mesh, scene + output.flush(); os.fsync(output.fileno()) + os.replace(temp, args.out) + report['verdict'] = 'PASS' + report['seconds'] = time.perf_counter()-started + print(json.dumps(report, ensure_ascii=False, allow_nan=False, indent=2)) + return 0 + + +if __name__ == '__main__': + raise SystemExit(main()) diff --git a/HM3DSem_exported/kits/unify_hm3dsem.py b/HM3DSem_exported/kits/unify_hm3dsem.py new file mode 100644 index 0000000000000000000000000000000000000000..b9969492a455a7228738d084ad33f2944954652b --- /dev/null +++ b/HM3DSem_exported/kits/unify_hm3dsem.py @@ -0,0 +1,38 @@ +#!/usr/bin/env python3 +"""Stream complete-scene JSONL into the published JSON array; fields are unchanged.""" +import argparse +import json +import os +from pathlib import Path +import time + + +def main(): + base = Path(__file__).resolve().parents[1] + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('--jsonl', type=Path, default=base/'hm3dsem.jsonl') + parser.add_argument('--out', type=Path, default=base/'multiroom/json_multi_merged.json') + args = parser.parse_args() + temp = args.out.with_suffix(args.out.suffix+'.part') + if args.jsonl.resolve() in {args.out.resolve(), temp.resolve()}: + parser.error('input must differ from output and scratch file') + started = time.perf_counter() + args.out.parent.mkdir(parents=True, exist_ok=True) + count = 0 + with args.jsonl.open(encoding='utf-8') as source, temp.open('w', encoding='utf-8') as output: + output.write('[\n') + for line in source: + if not line.strip(): + continue + if count: + output.write(',\n') + output.write(line.rstrip('\n')) + count += 1 + output.write('\n]\n') + output.flush(); os.fsync(output.fileno()) + os.replace(temp, args.out) + print(json.dumps({'scenes': count, 'seconds': time.perf_counter()-started, 'out': str(args.out)})) + + +if __name__ == '__main__': + main() diff --git a/HM3DSem_exported/kits/verify_hm3dsem.py b/HM3DSem_exported/kits/verify_hm3dsem.py new file mode 100644 index 0000000000000000000000000000000000000000..f76be9921d1ed1fedc2d16cd70f5847a2c4fa320 --- /dev/null +++ b/HM3DSem_exported/kits/verify_hm3dsem.py @@ -0,0 +1,336 @@ +#!/usr/bin/env python3 +"""Independent source acceptance gate. Does not import convert or unify. + +Re-decodes GLBs, welds structured vertex records, rebuilds the graph and component +extrema, and uses Qhull for room boundaries (converter: monotone chain). +Selection/null policies and source accounting are checked for every +selected scene. Small fixed numerical tolerances cover independent arithmetic, +never component-volume ties. Reports go to stdout; any failure exits nonzero. +""" +from __future__ import annotations + +import argparse +from collections import Counter, defaultdict +from contextlib import ExitStack +import csv +import io +import json +import math +from pathlib import Path +import struct +import sys +import tarfile +import time + +import numpy as np +from scipy.sparse import coo_matrix +from scipy.sparse.csgraph import connected_components +from scipy.spatial import ConvexHull + +BASE = Path(__file__).resolve().parents[1] +STRUCTURES = {'wall', 'floor', 'ceiling', 'shower wall', 'shower floor', 'shower ceiling'} +META = ('furnished navmesh_quality large_holes portals_blocked low_interactive_potential ' + 'doors_scanned_closed unfurnished num_floors overall_scene_quality diversity ' + 'num_rooms navigable_area navigation_complexity scene_clutter floor_space').split() + + +def strict_json(text): + def bad(value): + raise ValueError(f'invalid JSON number {value}') + return json.loads(text, parse_constant=bad) + + +def sources(data, stack): + result = {} + for split in ('train', 'val'): + name = f'hm3d-{split}-semantic-annots-v0.2.tar' + t = stack.enter_context(tarfile.open(data/name)) + members = [m for m in t.getmembers() if m.isfile()] + txt = {}; glb = {} + for m in members: + parts = m.name.split('/') + sid, basename = parts + stem = sid.split('-', 1)[1]+'.semantic.' + if basename == stem+'txt': txt[sid] = m + elif basename == stem+'glb': glb[sid] = m + else: raise ValueError('unexpected source member suffix') + with tarfile.open(data/f'hm3d-{split}-semantic-configs-v0.2.tar') as c: + config = json.load(c.extractfile(f'hm3d_annotated_{split}_basis.scene_dataset_config.json')) + expected = [x.split('/')[0] for x in config['stages']['paths']['.glb']] + if len(expected) != len(set(expected)) or set(txt) != set(glb) or set(txt) != set(expected): + raise ValueError('source scene accounting mismatch') + for sid in txt: + if sid in result: raise ValueError('scene occurs in two splits') + result[sid] = (t, name, txt[sid], glb[sid]) + with (data/'metadata.csv').open(newline='', encoding='utf-8') as f: + reader = csv.DictReader(f) + meta = {} + for row in reader: + meta[row['scene']] = row + return result, meta + + +def annotations(raw): + lines = raw.decode('utf-8').splitlines() + if not lines or lines[0] != 'HM3D Semantic Annotations': + raise ValueError('invalid annotation header') + records = [] + for cells in csv.reader(io.StringIO('\n'.join(lines[1:])), strict=True): + if not cells: continue + ident, color, category, region = cells + ident, region = int(ident), int(region) + int(color, 16) + if ident < 0 or region < 0 or not category.strip(): + raise ValueError('invalid source instance/region/category') + records.append((ident, color, category, region)) + if len({r[0] for r in records}) != len(records): + raise ValueError('duplicate source instance') + return sorted(records) + + +def decode_mesh(raw): + stream = io.BytesIO(raw) + if stream.read(4) != b'glTF' or struct.unpack(' len(binary) or end > length or stride < width*dtype.itemsize): + raise ValueError('source buffer overrun') + # Decode each component separately with a byte stride, not a shared reader. + out = np.empty((count, width), dtype=dtype) + for k in range(width): + out[:, k] = np.ndarray((count,), dtype=dtype, buffer=binary, + offset=start+offset+k*dtype.itemsize, strides=(stride,)) + return out + nodes = doc['nodes'] + roots = doc['scenes'][doc.get('scene', 0)]['nodes'] + mesh_ids = [] + for node in (nodes[i] for i in roots): + if node.get('children'): raise ValueError('unsupported hierarchy') + if 'mesh' in node: + if node.get('extensions') or any(k in node for k in ('matrix', 'rotation', 'translation', 'scale', 'skin', 'weights')): + raise ValueError('unsupported mesh transform') + mesh_ids.append(node['mesh']) + vv, cc, ff, offset = [], [], [], 0 + for mesh_id in mesh_ids: + for p in doc['meshes'][mesh_id]['primitives']: + if p.get('mode', 4) != 4 or p.get('targets') or p.get('extensions'): raise ValueError('unsupported primitive') + v = values(p['attributes']['POSITION'], {5126}, 'VEC3', False) + c = values(p['attributes']['COLOR_0'], {5123}, 'VEC4', True) + f = values(p['indices'], {5121, 5123, 5125}, 'SCALAR', False).ravel() + if len(v) != len(c) or len(f) % 3 or (len(f) and f.max() >= len(v)): raise ValueError('invalid source triangle') + vv.append(v); cc.append(c); ff.append(f.astype(np.int64).reshape(-1, 3)+offset) + offset += len(v) + if not vv: + return np.empty((0, 3)), np.empty((0, 4), dtype=np.uint16), np.empty((0, 3), dtype=np.int64) + return np.concatenate(vv).astype(np.float64), np.concatenate(cc), np.concatenate(ff) + + +def recompute_geometry(v, c, triangles, rows): + # A 65536-entry lookup independently expresses the fixed sRGB conversion. + linear = np.arange(65536, dtype=np.float64)/65535 + lut = np.empty_like(linear) + small = linear <= 0.0031308 + lut[small] = linear[small]*12.92 + lut[~small] = 1.055*linear[~small]**(1/2.4)-0.055 + lut = np.floor(lut*255+0.5).astype(np.uint32) + rgb = lut[c[:, :3]] + colors = rgb[:, 0]*65536+rgb[:, 1]*256+rgb[:, 2] + bad = set(map(int, colors[~np.isfinite(v).all(axis=1)])) + if bad - {int(r[1], 16) for r in rows}: raise ValueError('unlocalized invalid source vertices') + # Structured numeric fields preserve exact position+RGBA welding. + records = np.empty(len(v), dtype=[('xyz', '= 0 for x in volumes): raise ValueError('invalid source volumes') + maximum = max(volumes) + best = [box for box, volume in zip(boxes, volumes) if volume == maximum] + if any(not np.array_equal(np.concatenate(b), np.concatenate(best[0])) for b in best[1:]): return None + return best[0] + + +def boundary_of(points): + xy = np.unique(points[:, :2], axis=0) + if len(xy) < 3: return None + delta = xy-xy[0] + direction = delta[1] + if np.all(direction[0]*delta[:, 1]-direction[1]*delta[:, 0] == 0): return None + hull = xy[ConvexHull(xy).vertices] + # Compare cyclic rings with a canonical start and direction. + return hull.tolist() + + +def ring(points): + if points is None: return None + pts = [tuple(p) for p in points] + i = pts.index(min(pts)); forward = pts[i:]+pts[:i] + backward = [forward[0]]+list(reversed(forward[1:])) + return min(forward, backward) + + +def near(a, b): + if a is None or b is None: return a is b + return bool(np.allclose(a, b, rtol=1e-10, atol=1e-9)) + + +def verify_scene(scene, sid, rows, mesh, meta, locator, check, stats): + by_region, by_color = defaultdict(list), defaultdict(list) + for r in rows: + by_region[r[3]].append(r); by_color[int(r[1], 16)].append(r) + boxes, points, bad = recompute_geometry(*mesh, rows) + expected_prov = {'source_file': locator, 'source_value': {k: meta[k] for k in META} if meta else None} + check(scene['provenance'] == expected_prov, 'provenance', sid) + check(scene['source_subset'] == ((meta['split'] or None) if meta else None), 'split', sid) + check(scene['scene_id'] == scene['sample_id'] == sid, 'scene_identity', sid) + check(scene['room_count'] == len(scene['rooms']) == len(by_region), 'room_count', sid) + check([r['room_id'] for r in scene['rooms']] == [f'{sid}::region_{k}' for k in sorted(by_region)], 'room_identity_order', sid) + known = [r['room_area_m2'] for r in scene['rooms'] if r['room_area_m2'] is not None] + check(scene['total_area'] == (sum(known) if known else None), 'total_area', sid) + for index, (region, original) in enumerate(sorted(by_region.items())): + if index >= len(scene['rooms']): break + room = scene['rooms'][index]; where = f'{sid}/{region}' + expected = [r for r in original if r[2] not in STRUCTURES] + structures = defaultdict(list) + for ident, color, category, _ in original: + if category in STRUCTURES: structures[category].append({'id': ident, 'color': color}) + check(room['source_fields'] == {'structural_instances': dict(sorted(structures.items()))}, 'structural_records', where) + check(room['room_index'] == index, 'room_index', where) + check(room['furniture_quantity'] == len(room['furniture']) == len(expected), 'furniture_quantity', where) + check([f['furniture_instance_id'] for f in room['furniture']] == [f'{sid}:{r[0]}' for r in expected], 'instance_identity_order', where) + for item, row in zip(room['furniture'], expected): + ident, color, category, _ = row; w = f'{sid}:{ident}'; rgb = int(color, 16) + check(item['furniture_category'] == category and item['source_fields'] == {'color': color}, 'instance_source', w) + box = chosen_box(boxes[rgb]) if len(by_color[rgb]) == 1 else None + if box is None: + check(item['furniture_position'] is None and item['furniture_size'] is None, 'expected_null_box', w) + reason = 'duplicate_color' if len(by_color[rgb]) > 1 else ('nonfinite' if rgb in bad else ('missing_vertices' if not boxes[rgb] else 'tied_boxes')) + stats['null_box_'+reason] += 1 + else: + low, high = box + check(item['furniture_position'] is not None and near([item['furniture_position'][k] for k in ('x','y','z')], (low+high)/2), 'box_center', w) + check(item['furniture_size'] is not None and near([item['furniture_size'][k] for k in ('width','length','height')], high-low), 'box_size', w) + stats['boxes_non_null'] += 1 + stats['boxes_degenerate'] += bool(np.any(high == low)) + for field, category in (('doors', 'door'), ('windows', 'window')): + check(room[field] == [{'furniture_index': i} for i, r in enumerate(expected) if r[2] == category], field, where) + surfaces = {} + for category in ('floor', 'ceiling'): + colors = {int(r[1], 16) for r in original if r[2] == category} + ambiguous = any(c in bad or len({(r[2], r[3]) for r in by_color[c]}) != 1 for c in colors) + arrays = [points[c] for c in sorted(colors) if c in points] + surfaces[category] = (np.vstack(arrays) if arrays else np.empty((0,3)), ambiguous) + floor, bad_floor = surfaces['floor']; ceiling, bad_ceiling = surfaces['ceiling'] + boundary = None if bad_floor else boundary_of(floor) + actual, want = ring(room['room_boundary']), ring(boundary) + check((actual is None and want is None) or (actual is not None and want is not None and len(actual) == len(want) and near(actual, want)), 'boundary', where) + if boundary is None: + area = None + else: + # Trapezoid form, independent of converter's cross-product sum. + area = abs(sum((a[0]+b[0])*(b[1]-a[1]) for a,b in zip(boundary,boundary[1:]+boundary[:1])))/2 + check(near(room['room_area_m2'], area), 'room_area', where) + height = None + if len(floor) and len(ceiling) and not (bad_floor or bad_ceiling): + candidate = float(max(ceiling[:,2])-min(floor[:,2])) + if math.isfinite(candidate) and candidate > 0: height = candidate + check(near(room['room_height'], height), 'room_height', where) + stats['boundary_null'] += boundary is None + stats['height_null'] += height is None + stats['room_type_null'] += room['room_type'] is None + stats.update(source_instances=len(rows), rooms=len(by_region), furniture=sum(r[2] not in STRUCTURES for r in rows), structural=sum(r[2] in STRUCTURES for r in rows)) + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('--data-dir', type=Path, default=BASE/'data') + parser.add_argument('--jsonl', type=Path, default=BASE/'hm3dsem.jsonl') + parser.add_argument('--merged', type=Path, default=BASE/'multiroom/json_multi_merged.json') + scope = parser.add_mutually_exclusive_group(required=True) + scope.add_argument('--limit', type=int) + scope.add_argument('--scene-id', action='append') + scope.add_argument('--all', action='store_true') + args = parser.parse_args() + if args.limit is not None and args.limit < 1: parser.error('--limit must be positive') + started = time.perf_counter(); failures = Counter(); examples = {}; stats = Counter(); timing = [] + def check(ok, code, where): + if not ok: + failures[code] += 1; examples.setdefault(code, where) + merged = strict_json(args.merged.read_text(encoding='utf-8')) + with ExitStack() as stack: + index, metadata = sources(args.data_dir, stack) + selected = sorted(set(args.scene_id)) if args.scene_id else sorted(index)[:args.limit] + if not selected or set(selected)-set(index): raise ValueError('empty or unknown source selection') + seen = [] + with args.jsonl.open(encoding='utf-8') as f: + for line in f: + if not line.strip(): continue + scene = strict_json(line) + n, sid = len(seen), scene['scene_id'] + seen.append(sid) + check(n < len(merged) and scene == merged[n], 'merged_record', sid) + if sid not in selected: + check(False, 'unexpected_scene', sid); continue + t, name, txt, glb = index[sid] + begin = time.perf_counter() + rows = annotations(t.extractfile(txt).read()) + mesh = decode_mesh(t.extractfile(glb).read()) + verify_scene(scene, sid, rows, mesh, metadata.get(sid), f'data/{name}::{txt.name}', check, stats) + timing.append({'scene_id': sid, 'seconds': time.perf_counter()-begin, 'vertices': len(mesh[0])}) + print(f'{sid}: verified in {timing[-1]["seconds"]:.3f}s', file=sys.stderr, flush=True) + del mesh + check(seen == selected, 'scene_accounting_order', f'expected {selected}; actual {seen}') + check(len(merged) == len(seen), 'merged_count', str(len(merged))) + report = {'verdict': 'FAIL' if failures else 'PASS', 'failures': dict(failures), 'first_example': examples, + 'counts': dict(stats), 'scenes': timing, 'seconds': time.perf_counter()-started} + print(json.dumps(report, indent=2, ensure_ascii=False, allow_nan=False)) + return 1 if failures else 0 + + +if __name__ == '__main__': + raise SystemExit(main())