Release ClickContrast

.gitattributes

@@ -59,3 +59,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
 data/train_pairs.jsonl filter=lfs diff=lfs merge=lfs -text
+data/all_pairs.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -14,14 +14,10 @@ size_categories:
 configs:
 - config_name: default
   data_files:
-  - split: train_pairs
-    path: data/train_pairs.jsonl
-  - split: train_replay
-    path: data/train_replay.jsonl
-  - split: refinement_pairs
-    path: data/refinement_pairs.jsonl
-  - split: refinement_replay
-    path: data/refinement_replay.jsonl
+  - split: all_pairs
+    path: data/all_pairs.jsonl
+  - split: all_replay
+    path: data/all_replay.jsonl
 ---
 
 # ClickContrast
@@ -31,17 +27,13 @@ It is the training dataset for [SelectGround-8B](https://huggingface.co/ruotian/
 
 | File | Rows | Supervision |
 |---|---:|---|
-| train_pairs | 3,790 | Target coordinate, target region, distractor region, candidate regions |
-| train_replay | 3,790 | Target coordinate |
-| refinement_pairs | 502 | Target coordinate, target region, distractor region, candidate regions |
-| refinement_replay | 502 | Target coordinate |
+| all_pairs | 4,292 | Target coordinate, target region, distractor region, candidate regions |
+| all_replay | 4,292 | Target coordinate |
 
-The initial training stage uses **all four files**, with a 1:1 pair-to-coverage sampling cycle.
-Refinement uses only the two refinement files, with two coverage microbatches per paired microbatch.
-The refinement rows are included in the 8,584-row total.
+Training uses `all_pairs` and `all_replay` with a 1:1 pair-to-coverage sampling cycle.
 
-Main pairs retain targets supported by both MAI-UI-8B and UI-Venus-1.5-8B, together with an incorrect Qwen3-VL-8B prediction on another parsed element outside a centered 2x target neighborhood.
-Refinement pairs focus on teacher-checked small controls and icons, using wrong-element predictions from a fine-tuned mining model.
+Pairs retain teacher-supported targets together with a model prediction on another parsed element outside a centered 2x target neighborhood.
+A focused subset covers teacher-checked small controls and icons.
 Coverage examples span 86 Click100k source categories.
 
 ## Format
@@ -66,7 +58,7 @@ The JSONL order is part of the frozen training recipe. The manifest records file
 
 ## Training
 
-Use the two-stage SFT-plus-selection-loss recipe in the [SelectGround code repository](https://github.com/zhangruotian/SelectGround).
+Use the SFT-plus-selection-loss recipe in the [SelectGround code repository](https://github.com/zhangruotian/SelectGround).
 All examples contribute coordinate loss. Only paired examples contribute selection loss.
 No external parser or teacher is needed to train from this release.
 

data/{train_pairs.jsonl → all_pairs.jsonl}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd0c46e4fa08c99db1861b46e9e0904872334b30013b73157529354a9daad6d6
-size 43795561
+oid sha256:106dbc5fe4d60e9a9cc428f1b5283e1542d3f93d782150770fa33727d167063c
+size 44080629

manifest.json

@@ -1,11 +1,11 @@
 {
   "counts": {
-    "train_pairs": 3790,
-    "train_replay": 3790,
-    "refinement_pairs": 502,
-    "refinement_replay": 502
+    "all_pairs": 4292,
+    "all_replay": 4292
   },
   "sha256": {
+    "data/all_pairs.jsonl": "106dbc5fe4d60e9a9cc428f1b5283e1542d3f93d782150770fa33727d167063c",
+    "data/all_replay.jsonl": "bfab80a36a56f994ee5aa317c476ad4d04ebe01b342fa76224a2938689c7f661",
     "images/fb02a8b7ac7339013370b5a30ff6b1eee72f709ac0e3e3b0904fc7fc0a8e0785.jpg": "fb02a8b7ac7339013370b5a30ff6b1eee72f709ac0e3e3b0904fc7fc0a8e0785",
     "images/9d44c847d1c649004980a842af69207da522bbf4b28458f2e8f8b8f625bbb61d.png": "9d44c847d1c649004980a842af69207da522bbf4b28458f2e8f8b8f625bbb61d",
     "images/178dd89d8410e4c5aea8e1e7deba19718dbf25572ebd1d919cf9de2e9c7020b0.jpg": "178dd89d8410e4c5aea8e1e7deba19718dbf25572ebd1d919cf9de2e9c7020b0",
@@ -3051,7 +3051,6 @@
     "images/543b9672c83845edd8ddb5447fded99f9036dc91b7d4ca817b869494e25ce4b8.png": "543b9672c83845edd8ddb5447fded99f9036dc91b7d4ca817b869494e25ce4b8",
     "images/833890c6eb338b13631da86cefa6fdbb717f4ebee77cc31ddfe4ffb9f7f8fcea.jpg": "833890c6eb338b13631da86cefa6fdbb717f4ebee77cc31ddfe4ffb9f7f8fcea",
     "images/c7b65eff301250ebd2d905650259fcbc104739425b1a5b78f1a7972123255f34.png": "c7b65eff301250ebd2d905650259fcbc104739425b1a5b78f1a7972123255f34",
-    "data/train_pairs.jsonl": "bd0c46e4fa08c99db1861b46e9e0904872334b30013b73157529354a9daad6d6",
     "images/5de20507c8c0a9fda6a3b03b0ad0f2d90d483e75584c57c2c860a3ce8324139b.png": "5de20507c8c0a9fda6a3b03b0ad0f2d90d483e75584c57c2c860a3ce8324139b",
     "images/5da0ad21f3ed286fe0b50737e120503245be34d610379154805af03aa58fb822.png": "5da0ad21f3ed286fe0b50737e120503245be34d610379154805af03aa58fb822",
     "images/ee352c7ea212e4acb311ecbf23b748ece39c23b7091d85013959caf014c9740f.jpg": "ee352c7ea212e4acb311ecbf23b748ece39c23b7091d85013959caf014c9740f",
@@ -6165,7 +6164,6 @@
     "images/af1585da15191614cddab0e6e277bf42d710dd13f53a821fd956a98a87585d64.jpg": "af1585da15191614cddab0e6e277bf42d710dd13f53a821fd956a98a87585d64",
     "images/d5b5e8e90c51f359012a30e9b898cba6f586d13c2a0e4883072b4c04cf1568c3.jpg": "d5b5e8e90c51f359012a30e9b898cba6f586d13c2a0e4883072b4c04cf1568c3",
     "images/cfbd7dc1be0ea644c349fb4876988637075e36d00388fb3642e834f2a32a9ca0.jpg": "cfbd7dc1be0ea644c349fb4876988637075e36d00388fb3642e834f2a32a9ca0",
-    "data/train_replay.jsonl": "d0342d2c62afe9710a7dd5e50378b2f2f1a706f035fe43cc7fc06dcb727a464b",
     "images/630494b1656128b70e40692b1455f540145b1e44979b55ad22355fc57c0714d7.png": "630494b1656128b70e40692b1455f540145b1e44979b55ad22355fc57c0714d7",
     "images/bde9b3b7ef58da6ce852d820d930a8a9cd7066d1e779f58e092fbd09cd93abe6.png": "bde9b3b7ef58da6ce852d820d930a8a9cd7066d1e779f58e092fbd09cd93abe6",
     "images/937a2a970e1f7f93c4abad9f920388955229978a5d768b0f46f87184568037e2.jpg": "937a2a970e1f7f93c4abad9f920388955229978a5d768b0f46f87184568037e2",
@@ -6403,7 +6401,6 @@
     "images/aa60ebcf27b50db3a1e1bc0201fc9e5b2dd321ad8959882f167e571214aa70fa.jpg": "aa60ebcf27b50db3a1e1bc0201fc9e5b2dd321ad8959882f167e571214aa70fa",
     "images/bc2df17410af4e8c9a162d3fbe64fd0ebf20e7e377314d5f44430978f6440182.jpg": "bc2df17410af4e8c9a162d3fbe64fd0ebf20e7e377314d5f44430978f6440182",
     "images/ff08552cb4609f7964bc8a3995ef3da9de2877383a6e63ed5b8a945af296f2a2.jpg": "ff08552cb4609f7964bc8a3995ef3da9de2877383a6e63ed5b8a945af296f2a2",
-    "data/refinement_pairs.jsonl": "e39ab8eaf2aac9957f989b8ad15914fd425680d4a38261ed2e0278b474284adb",
     "images/abbd149ffd959efffd7999fcbda157ae38621d2c6d1efaf55a02d17e15a66ea3.jpg": "abbd149ffd959efffd7999fcbda157ae38621d2c6d1efaf55a02d17e15a66ea3",
     "images/90287dfc6762f5b2785d5deaca013dcd53b35c95b496c719bbf685a1d2338c8d.jpg": "90287dfc6762f5b2785d5deaca013dcd53b35c95b496c719bbf685a1d2338c8d",
     "images/1527795dc2309f674ae05f7f0c331ac3453c2bd8e871c4fae5dbb47fd7c7d601.png": "1527795dc2309f674ae05f7f0c331ac3453c2bd8e871c4fae5dbb47fd7c7d601",
@@ -6698,7 +6695,6 @@
     "images/accbf02777dbe315101e788083ba093b5e35d489bcb2df0fa1b6c6437abac240.png": "accbf02777dbe315101e788083ba093b5e35d489bcb2df0fa1b6c6437abac240",
     "images/5786c3820ad4c6a0f33f7445adc9acf04c70b91ff25f89617279cb4c15f5b3ba.jpg": "5786c3820ad4c6a0f33f7445adc9acf04c70b91ff25f89617279cb4c15f5b3ba",
     "images/b68240fcb45156a6898449efca35e408e98212999910114b2ad2026496583047.jpg": "b68240fcb45156a6898449efca35e408e98212999910114b2ad2026496583047",
-    "images/7e523f332345b13ed4e9e89e73da7d96d66dce10f7ef38ac5cc3cb908d699aae.png": "7e523f332345b13ed4e9e89e73da7d96d66dce10f7ef38ac5cc3cb908d699aae",
-    "data/refinement_replay.jsonl": "3e971de615878c6b707a6a46fde5d2c78c3a289a2378f4d81e63a803f4b781cf"
+    "images/7e523f332345b13ed4e9e89e73da7d96d66dce10f7ef38ac5cc3cb908d699aae.png": "7e523f332345b13ed4e9e89e73da7d96d66dce10f7ef38ac5cc3cb908d699aae"
   }
 }