Upload folder: ESTP-Bench
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- .gitattributes +12 -0
- ESTP-Bench/.gitattributes +754 -0
- ESTP-Bench/.mdl +0 -0
- ESTP-Bench/.msc +3 -0
- ESTP-Bench/.mv +1 -0
- ESTP-Bench/estp_dataset/__pycache__/livechat.cpython-310.pyc +0 -0
- ESTP-Bench/estp_dataset/benchmark/__init__.py +2 -0
- ESTP-Bench/estp_dataset/benchmark/__pycache__/__init__.cpython-310.pyc +0 -0
- ESTP-Bench/estp_dataset/benchmark/__pycache__/benchmark.cpython-310.pyc +0 -0
- ESTP-Bench/estp_dataset/benchmark/__pycache__/estp.cpython-310.pyc +0 -0
- ESTP-Bench/estp_dataset/benchmark/benchmark.py +0 -0
- ESTP-Bench/estp_dataset/benchmark/check_error_data.ipynb +97 -0
- ESTP-Bench/estp_dataset/benchmark/estp.py +607 -0
- ESTP-Bench/estp_dataset/benchmark/eval.py +428 -0
- ESTP-Bench/estp_dataset/benchmark/eval_cost.py +170 -0
- ESTP-Bench/estp_dataset/benchmark/eval_cqa.py +347 -0
- ESTP-Bench/estp_dataset/benchmark/eval_findcase.py +446 -0
- ESTP-Bench/estp_dataset/benchmark/eval_singleQA.sh +387 -0
- ESTP-Bench/estp_dataset/benchmark/evalate_singleQA.py +484 -0
- ESTP-Bench/estp_dataset/benchmark/merge_prediction_result.ipynb +44 -0
- ESTP-Bench/estp_dataset/cqa_anno.json +0 -0
- ESTP-Bench/estp_dataset/dataset.py +1 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2.json +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part0 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part1 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part2 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part3 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part4 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part5 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part6 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part7 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2.json +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part0 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part1 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part2 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part3 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part4 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part5 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part6 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part7 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part0 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part1 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part0 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part1 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part2 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part3 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part4 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part5 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part6 +0 -0
- ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part7 +0 -0
.gitattributes
CHANGED
|
@@ -57,3 +57,15 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 57 |
# Video files - compressed
|
| 58 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 59 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
# Video files - compressed
|
| 58 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 59 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
| 60 |
+
ESTP-Bench/.msc filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
ESTP-Bench/estp_dataset/estpCqa_ours/LivebaseStage2.5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
ESTP-Bench/estp_dataset/estpCqa_ours/LivebaseStage2.5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
ESTP-Bench/estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
ESTP-Bench/estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
ESTP-Bench/estp_dataset/estpSqa_ours/LIVE_IT0.95.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 66 |
+
ESTP-Bench/estp_dataset/estp_bench_sq_VideollmOnline0.9.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 67 |
+
ESTP-Bench/estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 68 |
+
ESTP-Bench/estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
ESTP-Bench/estp_dataset/eval.ipynb filter=lfs diff=lfs merge=lfs -text
|
| 70 |
+
ESTP-Bench/estp_dataset/tmp_predict_minicpmv_contextual.json filter=lfs diff=lfs merge=lfs -text
|
| 71 |
+
ESTP-Bench/estp_dataset/tmp_predict_minicpmv_v2.json filter=lfs diff=lfs merge=lfs -text
|
ESTP-Bench/.gitattributes
ADDED
|
@@ -0,0 +1,754 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
| 2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
| 3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
| 4 |
+
*.bin.* filter=lfs diff=lfs merge=lfs -text
|
| 5 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
| 6 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
| 7 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
| 8 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 9 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
| 10 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
| 11 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
| 12 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
| 13 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
| 14 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
| 15 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
| 16 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
| 17 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
| 18 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
| 19 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
| 20 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
| 21 |
+
*.tar filter=lfs diff=lfs merge=lfs -text
|
| 22 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
| 23 |
+
*.mat filter=lfs diff=lfs merge=lfs -text
|
| 24 |
+
*.npz filter=lfs diff=lfs merge=lfs -text
|
| 25 |
+
*.npy filter=lfs diff=lfs merge=lfs -text
|
| 26 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
| 27 |
+
*.hdf5 filter=lfs diff=lfs merge=lfs -text
|
| 28 |
+
*.pickle filter=lfs diff=lfs merge=lfs -text
|
| 29 |
+
*.pkl filter=lfs diff=lfs merge=lfs -text
|
| 30 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
| 31 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
| 32 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
| 33 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
+
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
| 35 |
+
*.tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
|
| 37 |
+
*.ark* filter=lfs diff=lfs merge=lfs -text
|
| 38 |
+
**/*ckpt*data* filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
**/*ckpt*.meta filter=lfs diff=lfs merge=lfs -text
|
| 40 |
+
**/*ckpt*.index filter=lfs diff=lfs merge=lfs -text
|
| 41 |
+
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
| 42 |
+
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
| 43 |
+
*.jpg filter=lfs diff=lfs merge=lfs -text
|
| 44 |
+
*.png filter=lfs diff=lfs merge=lfs -text
|
| 45 |
+
*.jpeg filter=lfs diff=lfs merge=lfs -text
|
| 46 |
+
*.bmp filter=lfs diff=lfs merge=lfs -text
|
| 47 |
+
*.gif filter=lfs diff=lfs merge=lfs -text
|
| 48 |
+
*.webp filter=lfs diff=lfs merge=lfs -text
|
| 49 |
+
*.mp3 filter=lfs diff=lfs merge=lfs -text
|
| 50 |
+
*.wav filter=lfs diff=lfs merge=lfs -text
|
| 51 |
+
*.wma filter=lfs diff=lfs merge=lfs -text
|
| 52 |
+
|
| 53 |
+
*.ogg filter=lfs diff=lfs merge=lfs -text
|
| 54 |
+
*.m4a filter=lfs diff=lfs merge=lfs -text
|
| 55 |
+
*.m3u8 filter=lfs diff=lfs merge=lfs -text
|
| 56 |
+
*.amr filter=lfs diff=lfs merge=lfs -text
|
| 57 |
+
*.audio filter=lfs diff=lfs merge=lfs -text
|
| 58 |
+
|
| 59 |
+
*.flv filter=lfs diff=lfs merge=lfs -text
|
| 60 |
+
|
| 61 |
+
*.mpg filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
*.asf filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
*.mov filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
*.mpeg filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
*.3gp filter=lfs diff=lfs merge=lfs -text
|
| 66 |
+
*.wmv filter=lfs diff=lfs merge=lfs -text
|
| 67 |
+
*.rmvb filter=lfs diff=lfs merge=lfs -text
|
| 68 |
+
*.rm filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
*.ts filter=lfs diff=lfs merge=lfs -text
|
| 70 |
+
*.mkv filter=lfs diff=lfs merge=lfs -text
|
| 71 |
+
*.flash filter=lfs diff=lfs merge=lfs -text
|
| 72 |
+
*.vob filter=lfs diff=lfs merge=lfs -text
|
| 73 |
+
*.pdf filter=lfs diff=lfs merge=lfs -text
|
| 74 |
+
*.ost filter=lfs diff=lfs merge=lfs -text
|
| 75 |
+
*.pst filter=lfs diff=lfs merge=lfs -text
|
| 76 |
+
*.doc filter=lfs diff=lfs merge=lfs -text
|
| 77 |
+
*.docx filter=lfs diff=lfs merge=lfs -text
|
| 78 |
+
*.txt filter=lfs diff=lfs merge=lfs -text
|
| 79 |
+
*.ppt filter=lfs diff=lfs merge=lfs -text
|
| 80 |
+
*.pptx filter=lfs diff=lfs merge=lfs -text
|
| 81 |
+
*.xls filter=lfs diff=lfs merge=lfs -text
|
| 82 |
+
*.xlsx filter=lfs diff=lfs merge=lfs -text
|
| 83 |
+
*.vsd filter=lfs diff=lfs merge=lfs -text
|
| 84 |
+
*.vsdx filter=lfs diff=lfs merge=lfs -text
|
| 85 |
+
*.jsonl filter=lfs diff=lfs merge=lfs -text
|
| 86 |
+
|
| 87 |
+
dataset_infos.json ignore
|
| 88 |
+
*.csv filter=lfs diff=lfs merge=lfs -text
|
| 89 |
+
*.tsv filter=lfs diff=lfs merge=lfs -text
|
| 90 |
+
|
| 91 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 92 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 93 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 94 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 95 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 96 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 97 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 98 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 99 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 100 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 101 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 102 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 103 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 104 |
+
estp_dataset/estpCqa_baseline/LLaVANextVideo7B_passive_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 105 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 106 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 107 |
+
estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 108 |
+
estp_dataset/estpCqa_baseline/CLIP_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 109 |
+
estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 110 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 111 |
+
estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 112 |
+
estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 113 |
+
estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 114 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 115 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 116 |
+
estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 117 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_debug0.175.json filter=lfs diff=lfs merge=lfs -text
|
| 118 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 119 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 120 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_passive_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 121 |
+
estp_dataset/estpCqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 122 |
+
estp_dataset/estpCqa_baseline/Lavila_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 123 |
+
estp_dataset/estpCqa_baseline/Lavila_streaming_v2evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 124 |
+
estp_dataset/estpCqa_baseline/MMDuet.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 125 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 126 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 127 |
+
estp_dataset/estpCqa_baseline/MMDuet.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 128 |
+
estp_dataset/estpCqa_baseline/MMDuet.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 129 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 130 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 131 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 132 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 133 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 134 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 135 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 136 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 137 |
+
estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 138 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 139 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 140 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 141 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 142 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 143 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 144 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 145 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 146 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 147 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passive_debug.json filter=lfs diff=lfs merge=lfs -text
|
| 148 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passive_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 149 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passiveevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 150 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passiveevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 151 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passiveevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 152 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passiveevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 153 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passive_v2evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 154 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 155 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 156 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 157 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 158 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 159 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 160 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 161 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_passive_debug.json filter=lfs diff=lfs merge=lfs -text
|
| 162 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_passive_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 163 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 164 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 165 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 166 |
+
estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 167 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 168 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 169 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 170 |
+
estp_dataset/estpCqa_baseline/MiniCPMV_passive_v2evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 171 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 172 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 173 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 174 |
+
estp_dataset/estpCqa_ours/LivebaseStage2.5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 175 |
+
estp_dataset/estpCqa_ours/LivebaseStage2.5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 176 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 177 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 178 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 179 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 180 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 181 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 182 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 183 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 184 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 185 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 186 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 187 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 188 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 189 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 190 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 191 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 192 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_5_5.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 193 |
+
estp_dataset/estpSqa_baseline/Lavila_streaming.json filter=lfs diff=lfs merge=lfs -text
|
| 194 |
+
estp_dataset/estpSqa_baseline/Lavila_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 195 |
+
estp_dataset/estpCqa_baseline/Qwen2VL_passive_evaluator_deepseek_1_2.json filter=lfs diff=lfs merge=lfs -text
|
| 196 |
+
estp_dataset/estpSqa_baseline/MMDuet.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 197 |
+
estp_dataset/estpSqa_baseline/MMDuet.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 198 |
+
estp_dataset/estpSqa_baseline/MMDuet.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 199 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 200 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 201 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 202 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 203 |
+
estp_dataset/estpSqa_baseline/MMDuet.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 204 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 205 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 206 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 207 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 208 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 209 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 210 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 211 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 212 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 213 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 214 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 215 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 216 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 217 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 218 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 219 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 220 |
+
estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 221 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 222 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 223 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 224 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 225 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 226 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 227 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 228 |
+
estp_dataset/estpCqa_ours/LivebaseStage2_lowevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 229 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 230 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 231 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 232 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 233 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 234 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 235 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 236 |
+
estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 237 |
+
estp_dataset/estpSqa_baseline/TimeChat_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 238 |
+
estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 239 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 240 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 241 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 242 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 243 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11_debug.json filter=lfs diff=lfs merge=lfs -text
|
| 244 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 245 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 246 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 247 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 248 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 249 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 250 |
+
estp_dataset/estpCqa_ours/LivebaseStage2_lowevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 251 |
+
estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 252 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 253 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 254 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 255 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_low.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 256 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 257 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_low.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 258 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_low.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 259 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_low.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 260 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_low.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 261 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 262 |
+
estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 263 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 264 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 265 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 266 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 267 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 268 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 269 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 270 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 271 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_lowevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 272 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 273 |
+
estp_dataset/estpCqa_ours/LivebaseStage3.5_highevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 274 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 275 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_low.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 276 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 277 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 278 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 279 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 280 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 281 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 282 |
+
estp_dataset/estpSqa_baseline/CLIP_streaming.json filter=lfs diff=lfs merge=lfs -text
|
| 283 |
+
estp_dataset/estpSqa_baseline/CLIP_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 284 |
+
estp_dataset/estpSqa_baseline/CLIP_streamingevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 285 |
+
estp_dataset/estpSqa_baseline/CLIP_streamingevaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 286 |
+
estp_dataset/estpSqa_baseline/CLIP_streamingevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 287 |
+
estp_dataset/estpSqa_baseline/EgoVLP_streaming.json filter=lfs diff=lfs merge=lfs -text
|
| 288 |
+
estp_dataset/estpSqa_baseline/EgoVLP_streamingevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 289 |
+
estp_dataset/estpSqa_baseline/EgoVLP_streaming_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 290 |
+
estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 291 |
+
estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 292 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 293 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 294 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 295 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passive.json filter=lfs diff=lfs merge=lfs -text
|
| 296 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 297 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 298 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_5_5.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 299 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 300 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 301 |
+
estp_dataset/estpSqa_baseline_5cases/LLaVANextVideo7B_fbf_5cases.json filter=lfs diff=lfs merge=lfs -text
|
| 302 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 303 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 304 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 305 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 306 |
+
estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_5_5.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 307 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 308 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_low.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 309 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 310 |
+
estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 311 |
+
estp_dataset/estpCqa_ours/LivebaseStage3_high0.31_1_low.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 312 |
+
estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 313 |
+
estp_dataset/estpSqa_baseline_5cases/LLaVAOneVision_fbf_5cases.json filter=lfs diff=lfs merge=lfs -text
|
| 314 |
+
|
| 315 |
+
estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5cases.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 316 |
+
estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_1_1.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 317 |
+
estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_5_5.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 318 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 319 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 320 |
+
estp_dataset/estpSqa_ours/LIVE_IT0.95.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 321 |
+
estp_dataset/eval.ipynb filter=lfs diff=lfs merge=lfs -text
|
| 322 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 323 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 324 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 325 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 326 |
+
estp_dataset/estpSqa_ours/LIVE_IT0.95.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 327 |
+
estp_dataset/estpSqa_ours/LIVE_IT0.95.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 328 |
+
estp_dataset/estpSqa_ours/LIVE_IT0.95.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 329 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 330 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 331 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 332 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 333 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json filter=lfs diff=lfs merge=lfs -text
|
| 334 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 335 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2_debug.json filter=lfs diff=lfs merge=lfs -text
|
| 336 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 337 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2_v2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 338 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2_v2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 339 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 340 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2_v2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 341 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2_v2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 342 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 343 |
+
estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 344 |
+
estp_dataset/estpSqa_ours/LivebaseStage2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 345 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 346 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 347 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 348 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 349 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 350 |
+
estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 351 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 352 |
+
estp_dataset/estpSqa_ours/LivebaseStage2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 353 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 354 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 355 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 356 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 357 |
+
estp_dataset/estpSqa_ours/LivebaseStage2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 358 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 359 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 360 |
+
estp_dataset/estpSqa_ours/LivebaseStage2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 361 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_11.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 362 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_11.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 363 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_11.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 364 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_low.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 365 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_low.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 366 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 367 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_low.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 368 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 369 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 370 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 371 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 372 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 373 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 374 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 375 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 376 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 377 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 378 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 379 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 380 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 381 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 382 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 383 |
+
estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 384 |
+
estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5casesevaluator_deepseek_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 385 |
+
estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5casesevaluator_deepseek_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 386 |
+
estp_dataset/estp_bench_sq.json filter=lfs diff=lfs merge=lfs -text
|
| 387 |
+
estp_dataset/estpSqa_ours/LivebaseStage3_high11_.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 388 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v3.json filter=lfs diff=lfs merge=lfs -text
|
| 389 |
+
estp_dataset/estp_bench_cq_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 390 |
+
estp_dataset/estp_bench_cq.json filter=lfs diff=lfs merge=lfs -text
|
| 391 |
+
estp_dataset/estp_bench_cq_v3.json filter=lfs diff=lfs merge=lfs -text
|
| 392 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame.json filter=lfs diff=lfs merge=lfs -text
|
| 393 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 394 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame55.json filter=lfs diff=lfs merge=lfs -text
|
| 395 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 396 |
+
estp_dataset/estp_bench_sq_VideollmOnline.json filter=lfs diff=lfs merge=lfs -text
|
| 397 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v4_fusion_dinov2.json filter=lfs diff=lfs merge=lfs -text
|
| 398 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v5_fusion_dinov2.json filter=lfs diff=lfs merge=lfs -text
|
| 399 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2.json filter=lfs diff=lfs merge=lfs -text
|
| 400 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 401 |
+
estp_dataset/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2_debug.json filter=lfs diff=lfs merge=lfs -text
|
| 402 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 403 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 404 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 405 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 406 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 407 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part5 filter=lfs diff=lfs merge=lfs -text
|
| 408 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_1_2.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 409 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part7 filter=lfs diff=lfs merge=lfs -text
|
| 410 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 411 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 412 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 413 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 414 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part6 filter=lfs diff=lfs merge=lfs -text
|
| 415 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json.part4 filter=lfs diff=lfs merge=lfs -text
|
| 416 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 417 |
+
estp_dataset/estp_bench_sq_VideollmOnline0.9.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 418 |
+
|
| 419 |
+
full_scale_2fps_max384/dedaa2ad-306d-47b3-a4c4-90b6b2c37e78.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 420 |
+
estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_llama_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 421 |
+
estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_llama_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 422 |
+
full_scale_2fps_max384/b3df1e5e-0c7a-4ac5-84b6-41acde78d57d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 423 |
+
full_scale_2fps_max384/b062ea19-1128-47fc-9ed2-da1644a62745.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 424 |
+
full_scale_2fps_max384/b493e8e3-4e18-4480-8e63-c21e25bf99f7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 425 |
+
full_scale_2fps_max384/b6d7c19a-c056-40d3-9f92-7b11c8c61d26.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 426 |
+
full_scale_2fps_max384/b4068f40-69e9-4217-81ba-84e59c25f125.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 427 |
+
full_scale_2fps_max384/ab93e55b-11cb-4332-b247-b3fb2fc67f53.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 428 |
+
full_scale_2fps_max384/b68f87f0-29be-4516-a34a-dda69dd84a56.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 429 |
+
full_scale_2fps_max384/af4828ab-2ff7-4935-87e0-08881989f178.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 430 |
+
full_scale_2fps_max384/b739682e-97d3-468c-ab32-eff051aaf9ec.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 431 |
+
full_scale_2fps_max384/b5b0a768-ff32-49e4-945f-dc31b32e72cc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 432 |
+
full_scale_2fps_max384/b2a1b8ca-99d6-4f26-953f-426e89649e90.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 433 |
+
full_scale_2fps_max384/b6b4f7a1-f3ec-425c-93eb-acc931acb2b5.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 434 |
+
full_scale_2fps_max384/b7682631-4cb0-4a23-81bf-597a91c90591.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 435 |
+
full_scale_2fps_max384/ab6f68a2-e5ba-48ce-ac24-7fb3df0584f2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 436 |
+
full_scale_2fps_max384/afbab8ce-797a-4c6e-800b-5acd9bf1653e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 437 |
+
full_scale_2fps_max384/bb03ea40-924a-450f-8521-9773c017395a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 438 |
+
full_scale_2fps_max384/b0a30641-ad93-4d6a-92a9-f277c78f1aa5.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 439 |
+
full_scale_2fps_max384/bd970f5b-3fd9-4ae9-9f2b-738e8ca54c1e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 440 |
+
full_scale_2fps_max384/c07e1c6f-3bf8-4d6b-9272-87c4b98f8336.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 441 |
+
full_scale_2fps_max384/bd1d22e0-2898-424f-8ec8-d47cab2e9fcb.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 442 |
+
full_scale_2fps_max384/b9eed644-56a9-4a0b-b1a3-5cc0d6688297.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 443 |
+
full_scale_2fps_max384/bc95b190-02a0-435a-a7cf-90d20ae0e23a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 444 |
+
full_scale_2fps_max384/c5938442-90c1-413c-be06-186222d8274f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 445 |
+
full_scale_2fps_max384/bb2297d3-91b7-4ec4-b39b-cc64694929db.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 446 |
+
full_scale_2fps_max384/b999ac9b-d285-4f44-944a-a0795969ac51.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 447 |
+
full_scale_2fps_max384/c517f92a-3b49-42f9-9927-c290a61f9202.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 448 |
+
full_scale_2fps_max384/c1344588-b8e2-409f-8c84-2d27a0fe7934.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 449 |
+
full_scale_2fps_max384/c037be7a-e56b-4053-a48b-c593bf069ef4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 450 |
+
full_scale_2fps_max384/c2241ea5-5c32-488e-b8d0-09676ef304ee.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 451 |
+
full_scale_2fps_max384/c6047868-51a3-4bb8-b833-5453c1fa563c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 452 |
+
full_scale_2fps_max384/c5fd91c3-07d2-4514-a693-381bca377a0b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 453 |
+
full_scale_2fps_max384/c745f5ed-8c4f-4736-a38c-1cd22ce5094a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 454 |
+
full_scale_2fps_max384/c4333895-ed19-42fe-9323-271a41bdfe4c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 455 |
+
full_scale_2fps_max384/c730a60b-236c-4e7b-9d7a-9967ff7ee9ac.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 456 |
+
full_scale_2fps_max384/c2d06df7-5d3a-4116-9edb-f1c81a4f669b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 457 |
+
full_scale_2fps_max384/c77015b4-40e1-44f5-b411-6b962dfd3157.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 458 |
+
full_scale_2fps_max384/c7d5d40f-840c-4be0-b79d-ab41394479a2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 459 |
+
full_scale_2fps_max384/c34effc7-6198-4c85-80c7-c8be85b77dd3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 460 |
+
full_scale_2fps_max384/c6e14005-854e-462f-a5a0-2474b759b1cc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 461 |
+
full_scale_2fps_max384/cbfeb6c8-ec1f-4884-bcbf-14d7764c4a36.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 462 |
+
full_scale_2fps_max384/c864505c-9d49-4da4-bf7e-254b4c348c03.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 463 |
+
full_scale_2fps_max384/989f3c2c-edd2-4f30-9925-56771e2bf7cc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 464 |
+
full_scale_2fps_max384/cc575a16-64fd-4cda-9248-5d85f506fdfd.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 465 |
+
full_scale_2fps_max384/c615d305-14fe-41cd-9099-caf420beb604.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 466 |
+
full_scale_2fps_max384/d651a4c5-b3ef-4f9a-b69c-5e9baebae9ad.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 467 |
+
full_scale_2fps_max384/d74c95ba-e072-44f8-bb2f-2631379af06d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 468 |
+
full_scale_2fps_max384/c628c1d2-4ce3-4f28-93a4-cc9d3dd9de11.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 469 |
+
full_scale_2fps_max384/acc6839e-9d6d-46db-921b-51812834d3b2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 470 |
+
full_scale_2fps_max384/d2606d87-2a9b-48df-96b2-5df1a4457215.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 471 |
+
full_scale_2fps_max384/d7951b74-71fc-4dd7-a874-408088870e72.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 472 |
+
full_scale_2fps_max384/d02fe2b4-fd4a-4b8a-baac-e1c90379acb9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 473 |
+
full_scale_2fps_max384/d27c5c7e-db9c-48f4-bea5-9970078d352e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 474 |
+
full_scale_2fps_max384/da80cc0f-c003-4ef3-b5ca-b09d3b105a03.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 475 |
+
full_scale_2fps_max384/c833a9f6-c59e-4f4e-816f-77f1a32c6c05.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 476 |
+
full_scale_2fps_max384/dddce8ac-09b0-4b13-b0fa-eb18e1f27b21.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 477 |
+
full_scale_2fps_max384/ce90f6f8-a6a8-47b5-bb4c-6e12698164a3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 478 |
+
full_scale_2fps_max384/eb1b6e62-7197-4a75-bc39-c373e558fd97.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 479 |
+
full_scale_2fps_max384/e7cda49b-1c03-4d74-a759-c2eeb96465b4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 480 |
+
full_scale_2fps_max384/e7742d04-2dc5-4046-9d0c-d45c6fe74d25.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 481 |
+
full_scale_2fps_max384/e4d961f5-176f-4dde-864a-bb94523312a1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 482 |
+
full_scale_2fps_max384/e72082e8-f9e6-42ac-ac45-de30f9adee9d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 483 |
+
full_scale_2fps_max384/eb628b6d-001d-48ab-90ed-730e47f299b9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 484 |
+
full_scale_2fps_max384/d939955c-6db9-4406-8b7b-085494320bc7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 485 |
+
full_scale_2fps_max384/e8657b65-be92-401d-a8d8-2fa32cb861c0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 486 |
+
full_scale_2fps_max384/b1543ea3-e86b-4545-9195-f1bf83e02b41.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 487 |
+
full_scale_2fps_max384/7bb3ad7f-9bdd-4806-afd6-efb7aa7babbc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 488 |
+
full_scale_2fps_max384/ec143498-9db7-4ee7-a0cf-43538717e570.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 489 |
+
full_scale_2fps_max384/eb2c9739-dba0-479c-81b7-9dda51c4ac20.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 490 |
+
full_scale_2fps_max384/de8ea8a7-6bdb-4679-9b17-9dc6a79e45e7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 491 |
+
full_scale_2fps_max384/ec9f9e41-aa3d-4466-8da9-69bcf23ffba2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 492 |
+
full_scale_2fps_max384/edaaa1c1-ea36-4318-86b0-becc9b7de0f4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 493 |
+
full_scale_2fps_max384/ed60dcdb-b273-44e7-b5dc-f9527d7c403f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 494 |
+
full_scale_2fps_max384/ee5ed7ca-667d-4a31-bdc8-13a9b8008246.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 495 |
+
full_scale_2fps_max384/eb5cb2b0-59e6-45da-af1b-ba86c7ab0b54.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 496 |
+
full_scale_2fps_max384/f1b5ae1e-742c-4802-9e61-ff8aae83e12e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 497 |
+
full_scale_2fps_max384/ec41b9e1-dbf9-45c3-bf88-4b250873cad0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 498 |
+
full_scale_2fps_max384/e6f1accd-e18d-43ec-b854-e7633b15cbf7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 499 |
+
full_scale_2fps_max384/f20fb254-9648-40ea-a500-5493a1ad9eac.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 500 |
+
full_scale_2fps_max384/efd052ce-4af1-48f4-8546-c13428ee92ec.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 501 |
+
full_scale_2fps_max384/f26ca7c3-2dfe-4e74-be61-529635df723f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 502 |
+
full_scale_2fps_max384/f442873e-6f2b-493f-a6db-718cd20a732a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 503 |
+
full_scale_2fps_max384/f265c11a-5b65-49e2-b77a-8735f2aa9a48.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 504 |
+
full_scale_2fps_max384/ec344610-74f4-4765-9c3f-0837ef78055d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 505 |
+
full_scale_2fps_max384/f5984df0-dbe0-42d6-8e74-a0e1368f48e8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 506 |
+
full_scale_2fps_max384/b8592091-2f2c-4c37-a931-3c84e3eff3c7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 507 |
+
full_scale_2fps_max384/eec01ded-9bf6-4da4-b61c-ac6a022e8ce4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 508 |
+
full_scale_2fps_max384/c485fef8-8456-48e0-8ccd-d6cb0ba8c6f8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 509 |
+
estp_dataset/result/estp_bench_sq_VideollmOnlineevaluator_llama_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 510 |
+
full_scale_2fps_max384/f7866811-1030-4be5-af61-7f1dd797f36b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 511 |
+
full_scale_2fps_max384/f57c99dc-02c7-4d3a-a4dc-893e74d523b1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 512 |
+
full_scale_2fps_max384/f9ce56a3-3e9b-4edd-b133-c05eed792ec0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 513 |
+
full_scale_2fps_max384/8d4895eb-b2d1-43b8-90f1-e80fcc37797b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 514 |
+
full_scale_2fps_max384/f89e52da-01ee-4773-95ca-e8a5431ccd65.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 515 |
+
full_scale_2fps_max384/fb02682a-1e0e-49ee-a0f4-0df7f5fbe98a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 516 |
+
full_scale_2fps_max384/fbb89f93-2922-49ac-b127-dd20711deff2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 517 |
+
full_scale_2fps_max384/f12a83a9-77ee-495c-9f11-2d41137a0b1a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 518 |
+
full_scale_2fps_max384/f40e0f92-2250-46c9-98a0-8ccf23d164e0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 519 |
+
full_scale_2fps_max384/fb7cc35d-3272-44a4-b8f2-15cd24fa345b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 520 |
+
full_scale_2fps_max384/grp-056db3f1-f957-46c8-b16b-c8fce22e78f9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 521 |
+
full_scale_2fps_max384/grp-1982af6d-8f53-43a5-b81a-cda5a8099e6e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 522 |
+
full_scale_2fps_max384/d14d6c08-b80a-446a-988e-5b57a079097c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 523 |
+
full_scale_2fps_max384/grp-27825332-6946-4c1c-9b4f-e4de4f13b718.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 524 |
+
full_scale_2fps_max384/grp-1a9bb53f-01ae-4bd7-afa3-a92570679a7a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 525 |
+
estp_dataset/result/estp_bench_sq_VideollmOnlineevaluator_llama_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 526 |
+
full_scale_2fps_max384/grp-5251b41e-91b1-4b87-9a18-c631b1d17e2e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 527 |
+
full_scale_2fps_max384/grp-2bccee1b-0ade-47ad-8e15-ad6c00861540.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 528 |
+
full_scale_2fps_max384/grp-304735ba-6bf5-4d39-bcb5-0dabddb11d68.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 529 |
+
full_scale_2fps_max384/fa21a282-6099-45cd-a4a7-39daf7973ec0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 530 |
+
full_scale_2fps_max384/grp-47305621-3729-44cc-8f1f-814faf214c40.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 531 |
+
full_scale_2fps_max384/d20f6551-0272-4a69-8b02-da7ffcf29e0f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 532 |
+
full_scale_2fps_max384/fe0d5ffc-686e-4846-bce0-1a18e513bf95.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 533 |
+
full_scale_2fps_max384/ffb6dfc1-d2f9-45b1-8e25-2c7d0f32d635.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 534 |
+
full_scale_2fps_max384/grp-5b24c19c-0bde-46ce-a32e-418b5ffaa8a3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 535 |
+
full_scale_2fps_max384/grp-67759ea1-52a2-45ee-b129-f482532d83df.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 536 |
+
full_scale_2fps_max384/grp-9c5c9efc-608f-4fdf-9c29-2251a451c8f9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 537 |
+
full_scale_2fps_max384/f6a6c5d5-3fe0-42af-9ddb-fa1a63521c0d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 538 |
+
full_scale_2fps_max384/grp-93d50e4c-0a3b-430a-8267-01fbe5f302f4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 539 |
+
full_scale_2fps_max384/grp-8912a138-6c68-424d-8be5-59e0d4f5c173.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 540 |
+
full_scale_2fps_max384/grp-88bae242-7f3a-45d4-b129-5d69b1a1e15a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 541 |
+
full_scale_2fps_max384/grp-7fb63b81-0a4f-4e9e-906b-60d1935d53c7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 542 |
+
full_scale_2fps_max384/grp-79f47a60-f1e9-4232-88b8-a1836e7dfd30.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 543 |
+
full_scale_2fps_max384/grp-c56e7e04-8787-4df1-98c6-352076f61e53.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 544 |
+
estp_dataset/result/tmp_predict_minicpmv_contextual_frame_by_frame.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 545 |
+
full_scale_2fps_max384/grp-c56fa4ea-b7e1-4d2c-b1f3-b97bb6fe0c56.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 546 |
+
estp_dataset/result/tmp_predict_minicpmv_contextual_frame_by_frame.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 547 |
+
full_scale_2fps_max384/grp-a9c519a7-4776-42d6-bcf1-270f0d302843.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 548 |
+
full_scale_2fps_max384/grp-cfe4d1f3-20c1-45d9-8bd3-5a7a8e7443f4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 549 |
+
full_scale_2fps_max384/grp-b93ab731-52c3-43b2-9d99-4229663ba67c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 550 |
+
full_scale_2fps_max384/grp-cec778f9-9b54-4b67-b013-116378fd7a85.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 551 |
+
full_scale_2fps_max384/grp-3d7ccb44-b05d-4b67-baae-4e0f55d8307b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 552 |
+
full_scale_2fps_max384/grp-f248a4aa-dd14-42fb-91fc-ec9f6bc16f20.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 553 |
+
estp_dataset/result/tmp_predict_minicpmv_contextual_frame_by_frame.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 554 |
+
full_scale_2fps_max384/grp-d8f0089b-cac5-41f4-af66-e11d6f54a43c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 555 |
+
full_scale_2fps_max384/grp-84322f50-7fce-4d49-a04b-9ecd7afd4119.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 556 |
+
full_scale_2fps_max384/grp-719d9e89-4eb2-49ea-be14-dc2637dc303f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 557 |
+
estp_dataset/result/tmp_predict_minicpmv_contextual_frame_by_frameevaluator_llama_5_5.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 558 |
+
full_scale_2fps_max384/grp-51fc62f8-00f4-44e3-af9c-7ebb63da6c3d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 559 |
+
full_scale_2fps_max384/grp-e5fb613b-1c6a-4439-8735-6414a6344c76.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 560 |
+
full_scale_2fps_max384/grp-ebce88dd-4852-4506-9dcc-5f5798ce1cbf.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 561 |
+
estp_dataset/result/tmp_predict_minicpmv_contextual_frame_by_frameevaluator_llama_5_5.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 562 |
+
estp_dataset/tmp_predict_minicpmv_v3_correctness.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 563 |
+
full_scale_2fps_max384/f50acf72-0347-47b3-a0a7-9916311d5c3a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 564 |
+
estp_dataset/tmp_predict_minicpmv_v3.json filter=lfs diff=lfs merge=lfs -text
|
| 565 |
+
full_scale_2fps_max384/f0041c6c-06b2-4c3b-b8e6-934305b4226e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 566 |
+
estp_dataset/tmp_predict_minicpmv_v3.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 567 |
+
estp_dataset/tmp_predict_minicpmv_v2.json filter=lfs diff=lfs merge=lfs -text
|
| 568 |
+
full_scale_2fps_max384/f9b214ac-cb2c-4a1e-b6f1-518293398e35.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 569 |
+
estp_dataset/tmp_predict_minicpmv_v3.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 570 |
+
estp_dataset/tmp_predict_minicpmv_v3_correctness.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 571 |
+
full_scale_2fps_max384/002c3b5c-ed86-4af3-99a1-4b497b7c8a86.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 572 |
+
estp_dataset/train_it/estpBenchSq5Cases_fbf_EWOStage2_v1.json filter=lfs diff=lfs merge=lfs -text
|
| 573 |
+
full_scale_2fps_max384/070776bf-f15a-4b6e-a906-d1ef9e132592.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 574 |
+
full_scale_2fps_max384/07309684-1f6e-4977-ab74-f3e63c361f36.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 575 |
+
full_scale_2fps_max384/029ff9ab-ac4c-43a5-ae53-153228c52a21.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 576 |
+
full_scale_2fps_max384/09b27569-7cb6-49e7-bf02-ec5333f04273.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 577 |
+
full_scale_2fps_max384/023bf95e-28de-43b4-a43f-720edba667a5.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 578 |
+
full_scale_2fps_max384/05b8b906-b9c5-48b4-be27-c91b5c28f92b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 579 |
+
full_scale_2fps_max384/0daaa66c-fb5f-4680-b84c-2f65f0f0bd9b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 580 |
+
full_scale_2fps_max384/0e6fb738-05fc-4dd5-9746-a8e10efe8c20.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 581 |
+
full_scale_2fps_max384/055ba601-c243-4d4a-8a2c-694a5c88824c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 582 |
+
full_scale_2fps_max384/10a04147-ae19-4c5e-b13b-b0b7005938a2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 583 |
+
estp_dataset/train_it/estpBenchSq5Cases_fbf_EWOLivebaseStage2_v1.json.part0 filter=lfs diff=lfs merge=lfs -text
|
| 584 |
+
full_scale_2fps_max384/10bf25b0-f781-4de3-b66c-9a4d06e3643a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 585 |
+
full_scale_2fps_max384/106e9447-8a09-4e89-964e-61c927827a56.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 586 |
+
full_scale_2fps_max384/0a01978c-e16d-4587-95f1-49efa3ab15d9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 587 |
+
estp_dataset/tmp_predict_minicpmv_contextual.json filter=lfs diff=lfs merge=lfs -text
|
| 588 |
+
full_scale_2fps_max384/0fcf23a0-fc53-4378-9a99-18c4f109f659.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 589 |
+
full_scale_2fps_max384/15ce76af-7eb0-4d27-9289-c9c71863f603.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 590 |
+
full_scale_2fps_max384/110352df-c7a3-4fec-a707-42936d101c28.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 591 |
+
full_scale_2fps_max384/134d9865-d9ee-47e5-a121-589a26c6b3ba.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 592 |
+
estp_dataset/train_it/estpBenchSq5Cases_fbf_EWOLivebaseStage2_v1.json.part2 filter=lfs diff=lfs merge=lfs -text
|
| 593 |
+
full_scale_2fps_max384/1a327b0c-b78b-4ae2-add0-49334bccddcd.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 594 |
+
full_scale_2fps_max384/182dffe9-d2bf-4a7d-b1d8-137470f0143d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 595 |
+
full_scale_2fps_max384/16d55886-6e1e-4195-9918-12dc4568320e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 596 |
+
full_scale_2fps_max384/13c76616-f168-4af0-8d2a-fe82ce232d6a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 597 |
+
estp_dataset/train_it/estpBenchSq5Cases_fbf_EWOLivebaseStage2_v1.json.part1 filter=lfs diff=lfs merge=lfs -text
|
| 598 |
+
estp_dataset/train_it/estpBenchSq5Cases_fbf_EWOLivebaseStage2_v1.json.part3 filter=lfs diff=lfs merge=lfs -text
|
| 599 |
+
full_scale_2fps_max384/174bee0b-b5d6-4e22-b81d-048005ae6ba6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 600 |
+
full_scale_2fps_max384/214b62df-2fb1-4b0e-af9d-42b0dd37defb.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 601 |
+
full_scale_2fps_max384/228eb02a-6c89-4aa0-9cd4-0cdab1550c83.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 602 |
+
full_scale_2fps_max384/1b333aaf-1c1b-4043-bbea-5b5eacbfaf5e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 603 |
+
full_scale_2fps_max384/1b9f06a7-b26d-4c74-863e-1d4fa22bbc37.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 604 |
+
full_scale_2fps_max384/04fe8f4d-081e-437e-a56a-2d53b6233fc9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 605 |
+
estp_dataset/tmp_predict_minicpmv.json filter=lfs diff=lfs merge=lfs -text
|
| 606 |
+
full_scale_2fps_max384/26f1c77e-814e-4609-ad67-12447d1627e1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 607 |
+
full_scale_2fps_max384/27b7edc0-0b05-4a4a-9dd0-893ba585faf8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 608 |
+
full_scale_2fps_max384/0e0d6704-1c6c-4a62-bc97-cc55658cf8ac.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 609 |
+
full_scale_2fps_max384/26dc2fae-b3b5-4c4a-ba51-efc6bc740f52.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 610 |
+
full_scale_2fps_max384/277e72e5-013a-446d-8463-778f1c786105.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 611 |
+
full_scale_2fps_max384/44e3114f-8654-48ee-97c5-9506311ba188.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 612 |
+
full_scale_2fps_max384/18e84829-901a-414d-8a2b-d1d2b3244db7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 613 |
+
full_scale_2fps_max384/448472cf-8cbb-4aa9-ae72-99ef72adae87.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 614 |
+
full_scale_2fps_max384/46d00bf5-ed73-4e5f-84eb-9c880eec10d8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 615 |
+
full_scale_2fps_max384/1558e9f2-d7f3-4a23-9627-4240f506d7df.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 616 |
+
full_scale_2fps_max384/45dc74e1-c8dd-443a-a7a6-ca4215144e97.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 617 |
+
full_scale_2fps_max384/26718e04-3f7a-443d-a589-ca814f05b60e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 618 |
+
full_scale_2fps_max384/200a8f32-c44a-463d-ba39-0190a0bddf7a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 619 |
+
full_scale_2fps_max384/26b4e1c1-54c7-419f-b2d3-63ea52fd3540.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 620 |
+
full_scale_2fps_max384/49d20f86-b516-49be-a27b-a450955c9f46.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 621 |
+
full_scale_2fps_max384/47078bd8-6b58-458c-8c34-8d905e21079a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 622 |
+
full_scale_2fps_max384/52d7e473-06a6-4464-81f4-08199bf5cb6a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 623 |
+
full_scale_2fps_max384/54337e20-df12-4133-9e34-618e0381265e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 624 |
+
full_scale_2fps_max384/4be2455a-6719-45d4-85c1-86de5bebdd44.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 625 |
+
full_scale_2fps_max384/51378d03-1fb8-481f-8e40-fb947567d25d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 626 |
+
full_scale_2fps_max384/434a5651-9901-4b32-b782-d9a2bb52e805.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 627 |
+
full_scale_2fps_max384/444a032d-faef-447e-a5dd-c3c8386672b6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 628 |
+
full_scale_2fps_max384/5093842e-7cf9-4509-bf04-6e0ec6b75b27.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 629 |
+
full_scale_2fps_max384/4ce9f93b-f44c-46fe-9050-3452d2f39abb.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 630 |
+
full_scale_2fps_max384/4c45d8f4-e026-4ca3-a1e4-6d32da434469.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 631 |
+
full_scale_2fps_max384/60209262-ab8e-4214-a8dd-4764a9a4b26d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 632 |
+
full_scale_2fps_max384/4a58fe42-9ebe-4dfa-a103-e577a151a3f2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 633 |
+
full_scale_2fps_max384/3d3cddf9-7c61-43af-8b1a-46aa78983a5a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 634 |
+
full_scale_2fps_max384/0ca23a40-6daf-4503-bfa2-f315a79b7317.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 635 |
+
full_scale_2fps_max384/484d4760-a55f-4f58-be67-e98d3920f1d3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 636 |
+
full_scale_2fps_max384/5a10fa29-fd9d-4d5f-baf3-d1bdffb77e4a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 637 |
+
full_scale_2fps_max384/4d770048-cfde-411f-ad53-e4d57a7dd0d1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 638 |
+
full_scale_2fps_max384/4c6f009a-f0ca-4557-8723-4111761e0a9a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 639 |
+
full_scale_2fps_max384/5c172db2-eca2-4f9a-8963-b2111123fef2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 640 |
+
full_scale_2fps_max384/5825062f-4827-407c-bbd3-91ca010d25c2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 641 |
+
full_scale_2fps_max384/5aa8f84b-aea8-4e40-b710-82b93875e1ae.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 642 |
+
full_scale_2fps_max384/5f9cb028-fa02-47ef-845f-3b149a527e46.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 643 |
+
full_scale_2fps_max384/603a427f-9191-4ca4-a1b0-dd3c5e7fda70.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 644 |
+
full_scale_2fps_max384/6322a6f2-f271-4335-bf46-d428a5a58298.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 645 |
+
full_scale_2fps_max384/63f0f27d-645c-4efc-9162-6f0cb0353bc8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 646 |
+
full_scale_2fps_max384/62623424-f671-4380-b12c-1acdb04afeee.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 647 |
+
full_scale_2fps_max384/680e61d7-86e0-47cd-a6e1-83868aaaf3f0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 648 |
+
full_scale_2fps_max384/6af04762-9ccf-41aa-bbfb-48a443c7cec3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 649 |
+
full_scale_2fps_max384/707fb659-9f76-482b-b83b-e0bc1f090cf7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 650 |
+
full_scale_2fps_max384/50e55c8f-b939-4b5c-86b5-e0b53d199d27.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 651 |
+
full_scale_2fps_max384/6d3939f5-5636-43c7-94bb-86318392bd2d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 652 |
+
full_scale_2fps_max384/543ac6fc-9f77-4cdd-b7fe-6721cb037916.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 653 |
+
full_scale_2fps_max384/6d20e1a8-bb8e-49ff-aa5d-ec861e7f5432.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 654 |
+
full_scale_2fps_max384/77362fa3-e23a-4942-bd53-7d58b23b979b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 655 |
+
full_scale_2fps_max384/74d38b4a-6f22-43b3-b631-9234c05f27ab.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 656 |
+
full_scale_2fps_max384/67bc312a-3b49-4847-aa79-63ad7d3e4217.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 657 |
+
full_scale_2fps_max384/74e37c3b-9b5d-41ef-a806-61000cb6857c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 658 |
+
full_scale_2fps_max384/74dbf8d0-b10d-4993-831e-7618801d2d95.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 659 |
+
full_scale_2fps_max384/50c1df1c-e016-472a-b0ac-d297b0c82e87.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 660 |
+
full_scale_2fps_max384/623bc0af-7d4f-4b85-9263-2d04e016283d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 661 |
+
full_scale_2fps_max384/478d7fa4-b174-4266-b0ff-bd180ba0b806.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 662 |
+
full_scale_2fps_max384/28bc1ee7-b0c1-4f30-934a-0ab665779d90.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 663 |
+
full_scale_2fps_max384/04ac4c40-22fd-42aa-a7f0-ee597ffb7058.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 664 |
+
full_scale_2fps_max384/2876b375-e848-412c-8a6f-0664cbab6a33.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 665 |
+
full_scale_2fps_max384/2c0c6508-397f-4c48-aeb7-abc7a3cae8d1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 666 |
+
full_scale_2fps_max384/1c2da42f-8b03-4643-bbd2-f29e2fc86bf6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 667 |
+
full_scale_2fps_max384/5e64c89d-34ce-49a3-949a-dc90385cdbcc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 668 |
+
full_scale_2fps_max384/775acd8e-086f-48cf-adf3-c154f0d0bd2d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 669 |
+
full_scale_2fps_max384/2c6e772c-4681-44a1-966c-dd84c65183ad.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 670 |
+
full_scale_2fps_max384/2c56528c-5499-4307-bfd1-cbb2d26b1d8b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 671 |
+
full_scale_2fps_max384/2a2ff7db-5460-4296-a8a7-946ba628226d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 672 |
+
full_scale_2fps_max384/053ae0a6-937d-4bdc-8001-d3dfe7899811.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 673 |
+
full_scale_2fps_max384/3bd5bf35-d6ac-43b2-ab75-1558a37c8550.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 674 |
+
full_scale_2fps_max384/7af460ca-08ba-4a4e-92d2-5f3e612e3642.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 675 |
+
full_scale_2fps_max384/7bca86ab-7327-4a38-8d2f-0a668da52f5d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 676 |
+
full_scale_2fps_max384/892a25e2-df0c-4577-af57-b720e73654f6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 677 |
+
full_scale_2fps_max384/7ed8365c-37f5-4033-978a-2d3c1faa0e33.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 678 |
+
full_scale_2fps_max384/2dbdd409-0be0-447c-8ae8-3f107fe9af80.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 679 |
+
full_scale_2fps_max384/2d442f12-3595-4b2d-b8c5-583b65c0c49f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 680 |
+
full_scale_2fps_max384/2f46d1e6-2a85-4d46-b955-10c2eded661c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 681 |
+
full_scale_2fps_max384/2f23b607-f2e6-4f58-85d3-004c840bead2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 682 |
+
full_scale_2fps_max384/2c365418-7598-48e2-a2c7-6c1885a351f4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 683 |
+
full_scale_2fps_max384/ae2cd423-6d11-4b18-959c-9e6ad872249f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 684 |
+
full_scale_2fps_max384/82935e33-3b6e-4a9c-91be-9887ac0beab4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 685 |
+
full_scale_2fps_max384/8a6a3316-d682-4a76-81db-b244081765c9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 686 |
+
full_scale_2fps_max384/2d1bf98e-a133-4f63-bc1d-91f718078a52.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 687 |
+
full_scale_2fps_max384/33fced0b-4bf6-4989-9174-cda7bb51ecef.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 688 |
+
full_scale_2fps_max384/36c093d9-dbff-4fdd-bbc4-5302ae13415a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 689 |
+
full_scale_2fps_max384/7fc72743-4b55-48de-b988-4c690bda912f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 690 |
+
full_scale_2fps_max384/89af735b-7c13-4366-befe-efd6cd5402ff.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 691 |
+
full_scale_2fps_max384/80111886-6bab-4b16-aac6-1dfa42357d8b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 692 |
+
full_scale_2fps_max384/81fc547e-75af-48cf-91c0-9f77ee633e1d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 693 |
+
full_scale_2fps_max384/3817b065-3d46-44ff-b6a6-8c4d646165f2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 694 |
+
full_scale_2fps_max384/89da8b4f-0b3a-4fba-8c77-58633d8b699b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 695 |
+
full_scale_2fps_max384/28151b38-7249-4c73-80b2-55c4b2046eba.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 696 |
+
full_scale_2fps_max384/84ccc9f7-8e02-4bfd-bd44-1658814df7cc.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 697 |
+
full_scale_2fps_max384/368e4468-1f75-4cdb-a9d0-9b8982a3cbef.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 698 |
+
full_scale_2fps_max384/89a95cd8-ca13-4edf-807e-d62138fd30e0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 699 |
+
full_scale_2fps_max384/272e8bcd-32a8-48ce-a9c4-a5f0a4f15145.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 700 |
+
full_scale_2fps_max384/8cda2388-6f45-46d4-8ded-4ba2e23b309f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 701 |
+
full_scale_2fps_max384/2e04b2bb-e5d3-4f66-a16e-f7756e14e749.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 702 |
+
full_scale_2fps_max384/85665a8d-6683-4aa0-bff8-3cd5e752c8f2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 703 |
+
full_scale_2fps_max384/3837916b-c0e8-4359-8a73-524d40660731.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 704 |
+
full_scale_2fps_max384/864fa3d8-9b18-44cb-a8e9-9b40765e2d0c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 705 |
+
full_scale_2fps_max384/78d7751d-36b9-473b-a1fb-9c14debbf61e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 706 |
+
full_scale_2fps_max384/7ddbf8a2-5b3e-44cd-b9dc-db17ec06831b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 707 |
+
full_scale_2fps_max384/8b82316c-c293-49a7-b401-a3cc026ab1d0.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 708 |
+
full_scale_2fps_max384/5315437a-2530-47e5-a0e4-4ac656c7cfa4.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 709 |
+
full_scale_2fps_max384/7913c8a8-1f99-4c2d-8a5d-9a5793162447.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 710 |
+
full_scale_2fps_max384/9531cee8-0f60-4a68-8243-397ab7670a75.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 711 |
+
full_scale_2fps_max384/3a481a46-928f-46fd-87e3-7725c5bcbbc9.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 712 |
+
full_scale_2fps_max384/8e84c762-b0f9-4a7d-982e-010f27999435.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 713 |
+
full_scale_2fps_max384/3af4cfae-ba88-42f8-ba7e-e832034dccba.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 714 |
+
full_scale_2fps_max384/712ef11e-dc19-4fe1-ae66-0d9ec934196e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 715 |
+
full_scale_2fps_max384/3dd00666-a34d-4b45-bb4c-7a4c2342e475.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 716 |
+
full_scale_2fps_max384/9089e01e-c28e-4767-9478-ffd7daba9b79.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 717 |
+
full_scale_2fps_max384/869a2290-5509-49d0-8cc6-e0a0230790d3.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 718 |
+
full_scale_2fps_max384/972f660f-27ad-49ae-bf00-8da9d6d8d708.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 719 |
+
full_scale_2fps_max384/95836c4a-e856-4c04-9c68-800b154b6bc6.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 720 |
+
full_scale_2fps_max384/9adc0c59-1167-4311-abf3-7b57a0743b1d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 721 |
+
full_scale_2fps_max384/7c969dc3-aea2-49d4-bd0e-48191b4e5a51.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 722 |
+
full_scale_2fps_max384/90e4acc5-28a2-4bd4-972c-c6f7e18e1cde.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 723 |
+
full_scale_2fps_max384/9b9364bc-a51c-43a1-a82e-628c4078d5c2.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 724 |
+
full_scale_2fps_max384/2b21d0f3-cb6b-4b5e-a418-fbe2dd69e023.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 725 |
+
full_scale_2fps_max384/8fe9742a-355f-4503-9652-0a9ebaf0ed42.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 726 |
+
full_scale_2fps_max384/98dc31a1-edbb-42d6-8d1c-0b43fe6f9c28.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 727 |
+
full_scale_2fps_max384/41309ada-0a81-47b2-bdd1-453a178db18a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 728 |
+
full_scale_2fps_max384/9e7fa594-5774-4681-a034-50a5c83cff8f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 729 |
+
full_scale_2fps_max384/a13a145f-920a-44ec-8aef-b489c097f4a7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 730 |
+
full_scale_2fps_max384/4220db3a-4708-4fd1-ab67-0ef1586906cd.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 731 |
+
full_scale_2fps_max384/960b820b-1807-4219-8e74-94c4d00abc41.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 732 |
+
full_scale_2fps_max384/984fadaf-0d6e-4814-ac10-00f37c71322f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 733 |
+
full_scale_2fps_max384/35b10335-0485-4d1f-8fa2-d4aeab8a022c.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 734 |
+
full_scale_2fps_max384/a4ec18ff-d6b4-4002-a5e7-83980534dc0e.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 735 |
+
full_scale_2fps_max384/4335b0cc-e10c-45df-aad3-a1fa66b7a7a8.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 736 |
+
full_scale_2fps_max384/863011c8-ebf7-4900-87d1-3c7930f95dab.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 737 |
+
full_scale_2fps_max384/a51026e3-4163-46ba-8d54-766fa8a17a7a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 738 |
+
full_scale_2fps_max384/a5bf18b5-15d4-4843-8269-3f31154e3736.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 739 |
+
full_scale_2fps_max384/42476ba0-f357-4bc2-ae65-8d2d92fd3faa.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 740 |
+
full_scale_2fps_max384/a7a31f23-8348-405e-a01a-cad63010bcc7.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 741 |
+
full_scale_2fps_max384/aa14c20d-cadf-4393-935f-52175427539d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 742 |
+
full_scale_2fps_max384/97c1c805-291f-4b2f-bef9-0495d9d2fc92.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 743 |
+
full_scale_2fps_max384/9e08a7b1-eb04-4d11-bb0b-f4cd42f0a819.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 744 |
+
full_scale_2fps_max384/4194c6fa-04c4-4206-a7a5-f3f9d6cb2cec.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 745 |
+
full_scale_2fps_max384/a9463083-1f17-40e2-bb71-cf4f16c4a54d.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 746 |
+
full_scale_2fps_max384/ab865129-78fa-47d4-8a50-ff8c5533246f.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 747 |
+
full_scale_2fps_max384/97eaf9ae-9b92-493d-a555-4fcdb7c039a1.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 748 |
+
full_scale_2fps_max384/a8486c50-4576-41e3-a9ad-ee809e40c648.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 749 |
+
full_scale_2fps_max384/39011e23-5fe4-41cd-b146-f3f9e3f3941a.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 750 |
+
full_scale_2fps_max384/aa6d234e-174f-4521-a404-03335c89f06b.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 751 |
+
full_scale_2fps_max384/ae11b3ee-0ef5-4ff8-90f6-c5fd3fef8072.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 752 |
+
full_scale_2fps_max384/a6213dda-1620-4d74-a2bf-5576b0c95e41.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 753 |
+
full_scale_2fps_max384/grp-0ee8d0ef-c5ba-42c5-9c40-8718faa081ec.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 754 |
+
full_scale_2fps_max384/3a158275-c4c2-4bde-a186-788341e43ed4.mp4 filter=lfs diff=lfs merge=lfs -text
|
ESTP-Bench/.mdl
ADDED
|
Binary file (42 Bytes). View file
|
|
|
ESTP-Bench/.msc
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e8c837d173da95685529158bc8b355b67d862a0f005f1e4fde528e532a7c0297
|
| 3 |
+
size 125749
|
ESTP-Bench/.mv
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
master
|
ESTP-Bench/estp_dataset/__pycache__/livechat.cpython-310.pyc
ADDED
|
Binary file (11.1 kB). View file
|
|
|
ESTP-Bench/estp_dataset/benchmark/__init__.py
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from .benchmark import *
|
| 2 |
+
from .estp import *
|
ESTP-Bench/estp_dataset/benchmark/__pycache__/__init__.cpython-310.pyc
ADDED
|
Binary file (196 Bytes). View file
|
|
|
ESTP-Bench/estp_dataset/benchmark/__pycache__/benchmark.cpython-310.pyc
ADDED
|
Binary file (155 Bytes). View file
|
|
|
ESTP-Bench/estp_dataset/benchmark/__pycache__/estp.cpython-310.pyc
ADDED
|
Binary file (14.7 kB). View file
|
|
|
ESTP-Bench/estp_dataset/benchmark/benchmark.py
ADDED
|
File without changes
|
ESTP-Bench/estp_dataset/benchmark/check_error_data.ipynb
ADDED
|
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cells": [
|
| 3 |
+
{
|
| 4 |
+
"cell_type": "code",
|
| 5 |
+
"execution_count": 6,
|
| 6 |
+
"metadata": {},
|
| 7 |
+
"outputs": [
|
| 8 |
+
{
|
| 9 |
+
"name": "stdout",
|
| 10 |
+
"output_type": "stream",
|
| 11 |
+
"text": [
|
| 12 |
+
"{'video_uid': 'd20f6551-0272-4a69-8b02-da7ffcf29e0f', 'conversation': [{'role': 'user', 'content': 'What should I do next to continue making a recipe after completing the earlier tasks: slice leek, heating a skillet, prepare soapy mixture in a bowl, peel garlic? Please provide clear instructions.', 'time': 434.8484}, {'role': 'assistant', 'content': 'First, add more water to the soapy mixture in the bowl to ensure it’s ready for cleaning.', 'time': 435.57229, 'start_time': 434.8484, 'end_time': 435.57229, 'stamp_time': 435.210345}, {'role': 'assistant', 'content': 'Next, dice the peeled garlic finely for better flavor distribution in your dish.', 'time': 445.22206, 'start_time': 435.57229, 'end_time': 445.22206, 'stamp_time': 440.397175}, {'role': 'assistant', 'content': 'Take a moment to check the recipe in the manual to confirm the next steps.', 'time': 839.977, 'start_time': 449.299, 'end_time': 839.977, 'stamp_time': 644.6379999999999}, {'role': 'assistant', 'content': 'Before proceeding, wash your hands thoroughly to maintain hygiene.', 'time': 860.10599, 'start_time': 840.78309, 'end_time': 860.10599, 'stamp_time': 850.44454}, {'role': 'assistant', 'content': 'Now, prepare the stock that will be used as a base for your soup.', 'time': 875.37843, 'start_time': 860.42941, 'end_time': 875.37843, 'stamp_time': 867.90392}, {'role': 'assistant', 'content': 'Add the prepared stock to the sliced leek in the skillet to start cooking the soup.', 'time': 987.484, 'start_time': 875.74605, 'end_time': 987.484, 'stamp_time': 931.6150250000001}, {'role': 'assistant', 'content': 'Clear the countertop to create more workspace and keep the area tidy.', 'time': 996.95592, 'start_time': 988.81578, 'end_time': 996.95592, 'stamp_time': 992.88585}, {'role': 'assistant', 'content': 'Wash the knife in the soapy mixture to clean it after use.', 'time': 1033.83901, 'start_time': 1006.96205, 'end_time': 1033.83901, 'stamp_time': 1020.4005299999999}, {'role': 'assistant', 'content': 'Arrange the plates into the cupboard to keep them ready for serving.', 'time': 1088.19204, 'start_time': 1068.89759, 'end_time': 1088.19204, 'stamp_time': 1078.544815}, {'role': 'assistant', 'content': 'Wash the kitchen utensils in the soapy mixture to ensure they’re clean and ready for use.', 'time': 1108.97613, 'start_time': 1095.386, 'end_time': 1108.97613, 'stamp_time': 1102.181065}, {'role': 'assistant', 'content': 'Finally, stir the potato leek soup in the skillet to combine all the ingredients evenly.', 'time': 1289.00491, 'start_time': 1109.72215, 'end_time': 1289.00491, 'stamp_time': 1199.36353}], 'duration': 867.9392953125, 'start_time': 434.8484, 'end_time': 1302.7876953125, 'Task Type': 'Action Reasoning', 'evaluator_output_text': [[], [], [], [], [], [], [], [], [], [], []], 'evaluator_output_reponse': [[], [], [], [], [], [], [], [], [], [], []]}\n",
|
| 13 |
+
"{'video_uid': 'grp-a9c519a7-4776-42d6-bcf1-270f0d302843', 'conversation': [{'role': 'user', 'content': 'What is the next thing I need to do to make pancakes after the steps already completed: gathers utensils on countertop, pours flour in bowl, stores flour in cabinet, removes ingredients from fridge, prepares flour mixture in bowl, turns on cooking gas, prepares flour mixture in bowl, inspects flour pack, heats butter in pan, prepares pancake in pan? Include any important details.', 'time': 381.307}, {'role': 'assistant', 'content': 'Next, fetch water in a bowl to add to your pancake mixture. This will help achieve the right consistency for the batter.', 'time': 382.23681, 'start_time': 381.307, 'end_time': 382.23681, 'stamp_time': 381.771905}, {'role': 'assistant', 'content': 'Now, inspect the pancake in the pan to ensure it’s cooking evenly and not burning. Adjust the heat if necessary.', 'time': 391.83298, 'start_time': 382.23681, 'end_time': 391.83298, 'stamp_time': 387.034895}, {'role': 'assistant', 'content': 'Remove any sauce or toppings from the fridge that you plan to serve with the pancakes.', 'time': 413.87677999999994, 'start_time': 393.33784, 'end_time': 413.87677999999994, 'stamp_time': 403.60731}, {'role': 'assistant', 'content': 'Declutter the kitchen countertop to create more space for plating and serving the pancakes.', 'time': 422.56539499999997, 'start_time': 414.07678, 'end_time': 422.56539499999997, 'stamp_time': 418.3210875}, {'role': 'assistant', 'content': 'Inspect the pancake in the pan again to check if it’s ready to flip or remove from the heat.', 'time': 446.946, 'start_time': 422.765395, 'end_time': 446.946, 'stamp_time': 434.8556975}, {'role': 'assistant', 'content': 'Clean a plate with a cloth to ensure it’s ready for serving the pancakes.', 'time': 503.40671499999996, 'start_time': 447.71982, 'end_time': 503.40671499999996, 'stamp_time': 475.5632675}, {'role': 'assistant', 'content': 'Prepare the next pancake in the pan, ensuring the batter is evenly spread for consistent cooking.', 'time': 508.35909, 'start_time': 503.606715, 'end_time': 508.35909, 'stamp_time': 505.9829025}, {'role': 'assistant', 'content': 'Wash any used utensils in the sink to keep your workspace clean and organized.', 'time': 592.967, 'start_time': 509.93854, 'end_time': 592.967, 'stamp_time': 551.45277}, {'role': 'assistant', 'content': 'Clean another plate with a cloth if needed, ensuring all serving dishes are ready.', 'time': 651.607, 'start_time': 592.96796, 'end_time': 651.607, 'stamp_time': 622.28748}, {'role': 'assistant', 'content': 'Prepare the next pancake in the pan, maintaining a steady heat for even cooking.', 'time': 659.0125, 'start_time': 654.49488, 'end_time': 659.0125, 'stamp_time': 656.75369}, {'role': 'assistant', 'content': 'Serve the pancake on a plate, arranging it neatly for presentation.', 'time': 742.45, 'start_time': 663.88021, 'end_time': 742.45, 'stamp_time': 703.165105}, {'role': 'assistant', 'content': 'Turn off the cooking gas to ensure safety once all pancakes are prepared.', 'time': 749.7833850000001, 'start_time': 742.5897, 'end_time': 749.7833850000001, 'stamp_time': 746.1865425000001}, {'role': 'assistant', 'content': 'Prepare the final pancake on the plate, adding any toppings or sauces as desired.', 'time': 752.07794, 'start_time': 749.9833850000001, 'end_time': 752.07794, 'stamp_time': 751.0306625000001}, {'role': 'assistant', 'content': 'Store any unused ingredients back in the fridge to keep them fresh for later use.', 'time': 855.817, 'start_time': 761.51526, 'end_time': 855.817, 'stamp_time': 808.6661300000001}, {'role': 'assistant', 'content': 'Cover the pancake with a lid to keep it warm while you finish up other tasks.', 'time': 869.93716, 'start_time': 857.18827, 'end_time': 869.93716, 'stamp_time': 863.562715}, {'role': 'assistant', 'content': 'Declutter the kitchen countertop once more to ensure a clean and tidy workspace.', 'time': 885.138505, 'start_time': 881.68724, 'end_time': 885.138505, 'stamp_time': 883.4128725}, {'role': 'assistant', 'content': 'Check the pancake on the plate to ensure it’s ready to serve and looks appetizing.', 'time': 1196.997, 'start_time': 885.338505, 'end_time': 1196.997, 'stamp_time': 1041.1677525}], 'duration': 831.6016588541665, 'start_time': 381.307, 'end_time': 1212.9086588541666, 'Task Type': 'Action Reasoning', 'evaluator_output_text': [[], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], []], 'evaluator_output_reponse': [[], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], []]}\n",
|
| 14 |
+
"{'video_uid': 'eb1b6e62-7197-4a75-bc39-c373e558fd97', 'conversation': [{'role': 'user', 'content': 'What are the sequential actions I need to take to accomplish my goal, like making dough balls?', 'time': 0.0}, {'role': 'assistant', 'content': 'To start, shape the dough into a ball by gently rolling it between your hands.', 'time': 29.20309, 'start_time': 0.0, 'end_time': 29.20309, 'stamp_time': 14.601545}, {'role': 'assistant', 'content': 'Continue shaping the dough into a smooth, round ball to ensure even cooking and texture.', 'time': 671.09652, 'start_time': 29.20309, 'end_time': 671.09652, 'stamp_time': 350.149805}], 'duration': 848.4666666666667, 'start_time': 0.0, 'end_time': 848.4666666666667, 'Task Type': 'Task Understanding', 'evaluator_output_text': [[], []], 'evaluator_output_reponse': [[], []]}\n",
|
| 15 |
+
"1212\n",
|
| 16 |
+
"1212\n"
|
| 17 |
+
]
|
| 18 |
+
}
|
| 19 |
+
],
|
| 20 |
+
"source": [
|
| 21 |
+
"import json,os\n",
|
| 22 |
+
"import numpy as np\n",
|
| 23 |
+
"\n",
|
| 24 |
+
"def load_multiple_json(file_path):\n",
|
| 25 |
+
" \"\"\"读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。\"\"\"\n",
|
| 26 |
+
" with open(file_path, 'r', encoding='utf-8') as f:\n",
|
| 27 |
+
" content = f.read()\n",
|
| 28 |
+
" \n",
|
| 29 |
+
" decoder = json.JSONDecoder()\n",
|
| 30 |
+
" pos = 0\n",
|
| 31 |
+
" results = []\n",
|
| 32 |
+
" content_length = len(content)\n",
|
| 33 |
+
" \n",
|
| 34 |
+
" while pos < content_length:\n",
|
| 35 |
+
" # 跳过空白字符\n",
|
| 36 |
+
" while pos < content_length and content[pos].isspace():\n",
|
| 37 |
+
" pos += 1\n",
|
| 38 |
+
" if pos >= content_length:\n",
|
| 39 |
+
" break\n",
|
| 40 |
+
" try:\n",
|
| 41 |
+
" obj, new_pos = decoder.raw_decode(content, pos)\n",
|
| 42 |
+
" results.append(obj)\n",
|
| 43 |
+
" pos = new_pos\n",
|
| 44 |
+
" except json.JSONDecodeError as e:\n",
|
| 45 |
+
" # 出现解析错误则退出循环\n",
|
| 46 |
+
" print(f\"JSON 解析错误: {e}\")\n",
|
| 47 |
+
" break\n",
|
| 48 |
+
" return results\n",
|
| 49 |
+
"\n",
|
| 50 |
+
"\n",
|
| 51 |
+
"eval_file1 = '/root/videollm-online/data/estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'\n",
|
| 52 |
+
"\n",
|
| 53 |
+
"parent_dir = os.path.dirname(eval_file1)\n",
|
| 54 |
+
"eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file1.split('/')[-1])]\n",
|
| 55 |
+
"eval_result1 = []\n",
|
| 56 |
+
"for eval_file in eval_files:\n",
|
| 57 |
+
" eval_result1 += load_multiple_json(eval_file)\n",
|
| 58 |
+
" \n",
|
| 59 |
+
" \n",
|
| 60 |
+
"eval_file2 = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_VideollmOnlineevaluator_llama_5_5.json'\n",
|
| 61 |
+
"\n",
|
| 62 |
+
"parent_dir = os.path.dirname(eval_file2)\n",
|
| 63 |
+
"eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file2.split('/')[-1])]\n",
|
| 64 |
+
"eval_result2 = []\n",
|
| 65 |
+
"for eval_file in eval_files:\n",
|
| 66 |
+
" eval_result2 += load_multiple_json(eval_file)\n",
|
| 67 |
+
" \n",
|
| 68 |
+
"for result in eval_result1:\n",
|
| 69 |
+
" if 'EWO' not in result:\n",
|
| 70 |
+
" print(result)\n",
|
| 71 |
+
"print(len(eval_result1))\n",
|
| 72 |
+
"print(len(eval_result2))\n"
|
| 73 |
+
]
|
| 74 |
+
}
|
| 75 |
+
],
|
| 76 |
+
"metadata": {
|
| 77 |
+
"kernelspec": {
|
| 78 |
+
"display_name": "videollm",
|
| 79 |
+
"language": "python",
|
| 80 |
+
"name": "python3"
|
| 81 |
+
},
|
| 82 |
+
"language_info": {
|
| 83 |
+
"codemirror_mode": {
|
| 84 |
+
"name": "ipython",
|
| 85 |
+
"version": 3
|
| 86 |
+
},
|
| 87 |
+
"file_extension": ".py",
|
| 88 |
+
"mimetype": "text/x-python",
|
| 89 |
+
"name": "python",
|
| 90 |
+
"nbconvert_exporter": "python",
|
| 91 |
+
"pygments_lexer": "ipython3",
|
| 92 |
+
"version": "3.10.14"
|
| 93 |
+
}
|
| 94 |
+
},
|
| 95 |
+
"nbformat": 4,
|
| 96 |
+
"nbformat_minor": 2
|
| 97 |
+
}
|
ESTP-Bench/estp_dataset/benchmark/estp.py
ADDED
|
@@ -0,0 +1,607 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import tqdm
|
| 2 |
+
import os
|
| 3 |
+
import time
|
| 4 |
+
import json
|
| 5 |
+
|
| 6 |
+
|
| 7 |
+
def list_user_query(conversation):
|
| 8 |
+
user_query = []
|
| 9 |
+
query_time = []
|
| 10 |
+
for i in range(len(conversation)):
|
| 11 |
+
if conversation[i]['role'].lower() == 'user':
|
| 12 |
+
user_query.append(conversation[i]['content'])
|
| 13 |
+
query_time.append(conversation[i]['time'])
|
| 14 |
+
# Sort user queries and query times by time
|
| 15 |
+
if len(user_query) > 1:
|
| 16 |
+
# Create pairs of (query, time) and sort by time
|
| 17 |
+
query_time_pairs = list(zip(user_query, query_time))
|
| 18 |
+
query_time_pairs.sort(key=lambda x: x[1])
|
| 19 |
+
|
| 20 |
+
# Unpack the sorted pairs back into separate lists
|
| 21 |
+
user_query = [pair[0] for pair in query_time_pairs]
|
| 22 |
+
query_time = [pair[1] for pair in query_time_pairs]
|
| 23 |
+
|
| 24 |
+
return user_query, query_time
|
| 25 |
+
|
| 26 |
+
def print_json(json_data):
|
| 27 |
+
print(json.dumps(json_data, indent=4))
|
| 28 |
+
|
| 29 |
+
PROMPT_TEMPLATE_PROACTIVE = '''You are an advanced image question-answering AI assistant. You have been provided with image and a question related to the images. Your task is to carefully analyze the images and provide the answer to the question. You need to carefully confirm whether the images content meet the conditions of the question, and then output the correct content.
|
| 30 |
+
|
| 31 |
+
Question: {}
|
| 32 |
+
|
| 33 |
+
The answer is:
|
| 34 |
+
'''
|
| 35 |
+
|
| 36 |
+
PROMPT_TEMPLATE_PASSIVE = '''You are an advanced video AI assistant. Given a video and a question, carefully analyze each frame of the video, identify all relevant moments that help answer the question, and provide the corresponding frame numbers along with the answer.
|
| 37 |
+
The format should be: '[frame idx] answer'. For example, [6] The object is a cup.
|
| 38 |
+
[60] The object is a cup.
|
| 39 |
+
[100] The object is a yellow cup.
|
| 40 |
+
|
| 41 |
+
Question: {}
|
| 42 |
+
|
| 43 |
+
The answer is:
|
| 44 |
+
'''
|
| 45 |
+
|
| 46 |
+
PROMPT_TEMPLATE_PASSIVE_GROUNDING = '''
|
| 47 |
+
Question: {}
|
| 48 |
+
|
| 49 |
+
'''
|
| 50 |
+
|
| 51 |
+
ONLINE_MODEL = ['VideollmOnline', 'MMDuet', 'EWO']
|
| 52 |
+
GROUNDING_MODEL = ['TimeChat']
|
| 53 |
+
STREAMING_MODEL = ['EgoVLP', 'CLIP', 'Lavila']
|
| 54 |
+
|
| 55 |
+
class ESTP_singleQ_benchmark:
|
| 56 |
+
def __init__(self, data, config=None):
|
| 57 |
+
self.data = data
|
| 58 |
+
task2number = {
|
| 59 |
+
"Object State Change Recognition": 0,
|
| 60 |
+
"Ego Object State Change Recognition": 0,
|
| 61 |
+
"Object Localization": 0,
|
| 62 |
+
"Action Recognition": 0,
|
| 63 |
+
"Action Reasoning": 0,
|
| 64 |
+
"Object Recognition": 0,
|
| 65 |
+
"Ego Object Localization": 0,
|
| 66 |
+
"Object Function": 0,
|
| 67 |
+
"Task Understanding": 0,
|
| 68 |
+
"Attribute Perception": 0,
|
| 69 |
+
"Information Function": 0,
|
| 70 |
+
"Text-Rich Understanding": 0
|
| 71 |
+
}
|
| 72 |
+
|
| 73 |
+
for k,v in data.items():
|
| 74 |
+
for kk,vv in v.items():
|
| 75 |
+
for ll in vv:
|
| 76 |
+
task2number[ll['Task Type'].strip()] += 1
|
| 77 |
+
|
| 78 |
+
print_json(task2number)
|
| 79 |
+
|
| 80 |
+
self.config = config
|
| 81 |
+
|
| 82 |
+
def eval(self, data, model, output_path, eval_mode):
|
| 83 |
+
"""
|
| 84 |
+
Evaluate the model on the given data and update the data with the model responses.
|
| 85 |
+
data: data input
|
| 86 |
+
model: model to evaluate
|
| 87 |
+
"""
|
| 88 |
+
video_root = self.config.video_root
|
| 89 |
+
# Create output directory if it doesn't exist
|
| 90 |
+
os.makedirs(os.path.dirname(output_path), exist_ok=True)
|
| 91 |
+
|
| 92 |
+
for k,v in tqdm.tqdm(list(data.items())):
|
| 93 |
+
for kk,vv in v.items():
|
| 94 |
+
for q_idx, qa in enumerate(vv):
|
| 95 |
+
if model.name() in qa.keys():
|
| 96 |
+
continue
|
| 97 |
+
video_path = os.path.join(video_root, k + '.mp4')
|
| 98 |
+
|
| 99 |
+
for conv in qa['conversation']:
|
| 100 |
+
if 'time' not in conv.keys():
|
| 101 |
+
conv['time'] = conv['start_time'] + (conv['end_time'] - conv['start_time']) / 2
|
| 102 |
+
qa['conversation'] = sorted(qa['conversation'], key=lambda x: x['time'])
|
| 103 |
+
|
| 104 |
+
start_time = qa['clip_start_time'] if 'clip_start_time' in qa.keys() else qa['start_time']
|
| 105 |
+
end_time = qa['clip_end_time'] if 'clip_end_time' in qa.keys() else qa['end_time']
|
| 106 |
+
max_time = qa['conversation'][-1]['end_time']
|
| 107 |
+
|
| 108 |
+
if 'question' in qa.keys():
|
| 109 |
+
inp = qa['question']
|
| 110 |
+
query_time = qa['question_time'] if 'question_time' in qa.keys() else start_time # start_time, end_time, question_time is all the in all video scale
|
| 111 |
+
else:
|
| 112 |
+
inp, query_time = list_user_query(qa['conversation'])
|
| 113 |
+
inp = inp[0]
|
| 114 |
+
query_time = query_time[0]
|
| 115 |
+
|
| 116 |
+
start_time = min(start_time, query_time)
|
| 117 |
+
end_time = min(end_time, max_time)
|
| 118 |
+
|
| 119 |
+
# if eval_mode == "frame_by_frame":
|
| 120 |
+
# dialog_history = eval_fbf(model, video_path, inp, start_time, end_time, query_time)
|
| 121 |
+
# elif eval_mode == "passive_inference":
|
| 122 |
+
# dialog_history = eval_passive_inference(model, video_path, inp, start_time, end_time, query_time)
|
| 123 |
+
# else:
|
| 124 |
+
# raise ValueError(f"Invalid eval mode: {eval_mode}")
|
| 125 |
+
|
| 126 |
+
try:
|
| 127 |
+
if eval_mode == "frame_by_frame":
|
| 128 |
+
dialog_history = eval_fbf(model, video_path, inp, start_time, end_time, query_time)
|
| 129 |
+
elif eval_mode == "passive_inference":
|
| 130 |
+
dialog_history = eval_passive_inference(model, video_path, inp, start_time, end_time, query_time)
|
| 131 |
+
else:
|
| 132 |
+
raise ValueError(f"Invalid eval mode: {eval_mode}")
|
| 133 |
+
except Exception as e:
|
| 134 |
+
print(f"Error {e} in {k} {kk} {q_idx}")
|
| 135 |
+
continue
|
| 136 |
+
qa[model.name()] = dialog_history
|
| 137 |
+
|
| 138 |
+
json.dump(data, open(output_path, 'w'), indent=4)
|
| 139 |
+
|
| 140 |
+
def eval_passive_inference(model, video_path, question, start_time, end_time, query_time=0):
|
| 141 |
+
"""
|
| 142 |
+
Evaluate the model by first feeding all video frames and then getting response
|
| 143 |
+
|
| 144 |
+
Args:
|
| 145 |
+
model: The model to be evaluated
|
| 146 |
+
video_path: The path to the video file
|
| 147 |
+
inp: The input question
|
| 148 |
+
start_time: The start time of the evaluation
|
| 149 |
+
end_time: The end time of the evaluation
|
| 150 |
+
query_time: The time when the question is asked
|
| 151 |
+
|
| 152 |
+
Returns:
|
| 153 |
+
dialog_history: List of conversation turns with timestamps
|
| 154 |
+
"""
|
| 155 |
+
|
| 156 |
+
if model.name() in ONLINE_MODEL:
|
| 157 |
+
dialog_history = model.Run(video_path, question, start_time, end_time, query_time)
|
| 158 |
+
elif model.name() in STREAMING_MODEL:
|
| 159 |
+
dialog_history = model.Run(video_path, question, start_time, end_time, query_time)
|
| 160 |
+
elif model.name() in GROUNDING_MODEL:
|
| 161 |
+
raise NotImplementedError(f"Grounding model {model.name()} is not implemented")
|
| 162 |
+
else:
|
| 163 |
+
query = question
|
| 164 |
+
inp = PROMPT_TEMPLATE_PASSIVE.format(query)
|
| 165 |
+
|
| 166 |
+
real_start_time = time.time()
|
| 167 |
+
response, frame_number = model.Run(video_path, inp, start_time, end_time)
|
| 168 |
+
real_end_time = time.time()
|
| 169 |
+
|
| 170 |
+
frame_fps = frame_number / (end_time - start_time)
|
| 171 |
+
|
| 172 |
+
# 2. parse response
|
| 173 |
+
response_list = response.split('\n')
|
| 174 |
+
response_list = [i.strip() for i in response_list if i.strip()]
|
| 175 |
+
answer_pairs = []
|
| 176 |
+
|
| 177 |
+
for response in response_list:
|
| 178 |
+
if '[' in response and ']' in response:
|
| 179 |
+
frame_idx = response.split('[')[1].split(']')[0]
|
| 180 |
+
if ',' in frame_idx:
|
| 181 |
+
frame_idxs = [int(x.strip()) for x in frame_idx.split(',')]
|
| 182 |
+
elif '-' in frame_idx:
|
| 183 |
+
# Handle range format like "1-7"
|
| 184 |
+
start, end = map(int, frame_idx.split('-'))
|
| 185 |
+
frame_idxs = [int((start+end) / 2)]
|
| 186 |
+
else:
|
| 187 |
+
try:
|
| 188 |
+
frame_idxs = [int(frame_idx)]
|
| 189 |
+
except:
|
| 190 |
+
continue
|
| 191 |
+
answer = response.split('[')[1].split(']')[1].strip()
|
| 192 |
+
for frame_idx in frame_idxs:
|
| 193 |
+
answer_time = start_time + frame_idx / frame_fps
|
| 194 |
+
answer_pairs.append((answer_time, answer))
|
| 195 |
+
|
| 196 |
+
# 3. generate dialog history
|
| 197 |
+
dialog_history = [{
|
| 198 |
+
'role': 'user',
|
| 199 |
+
'content': question,
|
| 200 |
+
'time': query_time,
|
| 201 |
+
'fps': frame_number / (real_end_time - real_start_time),
|
| 202 |
+
'cost': real_end_time - real_start_time
|
| 203 |
+
}]
|
| 204 |
+
|
| 205 |
+
for answer_time, answer in answer_pairs:
|
| 206 |
+
dialog_history.append({
|
| 207 |
+
'role': 'assistant',
|
| 208 |
+
'content': answer,
|
| 209 |
+
'time': answer_time,
|
| 210 |
+
'fps': frame_number / (real_end_time - real_start_time),
|
| 211 |
+
'cost': real_end_time - real_start_time
|
| 212 |
+
})
|
| 213 |
+
return dialog_history
|
| 214 |
+
|
| 215 |
+
def eval_streaming(model, video_path, question, start_time, end_time, query_time=0):
|
| 216 |
+
"""
|
| 217 |
+
Evaluate the model on the data streaming
|
| 218 |
+
"""
|
| 219 |
+
# TODO: implement this
|
| 220 |
+
dialog_history = model.Run(video_path, question, start_time, end_time, query_time)
|
| 221 |
+
return dialog_history
|
| 222 |
+
|
| 223 |
+
def eval_fbf(model, video_path, question, start_time, end_time, query_time=0):
|
| 224 |
+
"""
|
| 225 |
+
Evaluate the model on the data frame by frame
|
| 226 |
+
|
| 227 |
+
Args:
|
| 228 |
+
model: The model to be evaluated
|
| 229 |
+
video_path: The path to the video file
|
| 230 |
+
inp: The input data
|
| 231 |
+
start_time: The start time of the evaluation
|
| 232 |
+
end_time: The end time of the evaluation
|
| 233 |
+
|
| 234 |
+
Returns:
|
| 235 |
+
None
|
| 236 |
+
"""
|
| 237 |
+
|
| 238 |
+
if model.name() in ONLINE_MODEL:
|
| 239 |
+
dialog_history = model.Run(video_path, question, start_time, end_time, query_time)
|
| 240 |
+
elif model.name() in STREAMING_MODEL:
|
| 241 |
+
raise NotImplementedError(f"Streaming model {model.name()} is not implemented")
|
| 242 |
+
elif model.name() in GROUNDING_MODEL:
|
| 243 |
+
raise NotImplementedError(f"Grounding model {model.name()} is not implemented")
|
| 244 |
+
else:
|
| 245 |
+
# query = f"Is it the right time to answer the question \"{inp}\"? You need to answer yes or no first, and if yes, please answer the question."
|
| 246 |
+
query = f"Is it the right time to answer the question \"{question}\"? You need to answer yes or no."
|
| 247 |
+
first_inp = PROMPT_TEMPLATE_PROACTIVE.format(query)
|
| 248 |
+
|
| 249 |
+
yes_query = f"Please answer the question: \"{question}\""
|
| 250 |
+
yes_inp = PROMPT_TEMPLATE_PROACTIVE.format(yes_query)
|
| 251 |
+
|
| 252 |
+
|
| 253 |
+
dialog_history = []
|
| 254 |
+
frame_fps = model.frame_fps # turn ask fps
|
| 255 |
+
|
| 256 |
+
current_time = min(start_time + 1 / frame_fps, end_time)
|
| 257 |
+
timecosts = []
|
| 258 |
+
|
| 259 |
+
while current_time <= end_time:
|
| 260 |
+
real_start_time = time.time()
|
| 261 |
+
response, frame_number = model.Run(video_path, first_inp, start_time, current_time)
|
| 262 |
+
real_end_time = time.time()
|
| 263 |
+
timecosts.append(real_end_time - real_start_time)
|
| 264 |
+
|
| 265 |
+
if 'yes' in response.lower():
|
| 266 |
+
real_start_time = time.time()
|
| 267 |
+
response, frame_number = model.Run(video_path, yes_inp, start_time, current_time)
|
| 268 |
+
real_end_time = time.time()
|
| 269 |
+
timecosts.append(real_end_time - real_start_time)
|
| 270 |
+
|
| 271 |
+
fps = (current_time-start_time) * frame_fps / sum(timecosts)
|
| 272 |
+
|
| 273 |
+
dialog_history.append({
|
| 274 |
+
'role': 'user', 'content': query, 'time': current_time, 'fps': fps, 'cost': timecosts[-1]
|
| 275 |
+
})
|
| 276 |
+
dialog_history.append({
|
| 277 |
+
'role': 'assistant', 'content': response, 'time': current_time, 'fps': fps, 'cost': timecosts[-1]
|
| 278 |
+
})
|
| 279 |
+
current_time += 1 / frame_fps
|
| 280 |
+
dialog_history.append({
|
| 281 |
+
'role': 'fps', 'content': (end_time - start_time) / sum(timecosts),
|
| 282 |
+
})
|
| 283 |
+
return dialog_history
|
| 284 |
+
|
| 285 |
+
|
| 286 |
+
PROMPT_TEMPLATE_C_PROACTIVE = '''You are an advanced image question-answering AI assistant. You have been provided with image and a question related to the images. Your task is to carefully analyze the images, provided context and provided the answer to the question. You need to carefully confirm whether the images content meet the conditions of the question, and then output the correct content.
|
| 287 |
+
|
| 288 |
+
{}
|
| 289 |
+
|
| 290 |
+
Here is the question. Answer it and don't confuse it with the previous conversation.
|
| 291 |
+
Question: {}
|
| 292 |
+
|
| 293 |
+
|
| 294 |
+
The answer is:
|
| 295 |
+
'''
|
| 296 |
+
|
| 297 |
+
PROMPT_TEMPLATE_C_PASSIVE = '''You are an advanced video question-answering AI assistant. Given a video and a question related to the video, carefully analyze each frame of the video and the provided context, identify all relevant moments that help answer the question, and provide the corresponding frame numbers along with the answer.
|
| 298 |
+
The format should be: '[frame idx] answer'. For example, [6] The object is a cup.
|
| 299 |
+
[60] The object is a cup.
|
| 300 |
+
[100] The object is a yellow cup.
|
| 301 |
+
|
| 302 |
+
{}
|
| 303 |
+
|
| 304 |
+
Here is the question. Answer it and don't confuse it with the previous conversation.
|
| 305 |
+
Question: {}
|
| 306 |
+
|
| 307 |
+
'''
|
| 308 |
+
|
| 309 |
+
PROMPT_TEMPLATE_C_PASSIVE_GROUNDING = '''
|
| 310 |
+
|
| 311 |
+
{}
|
| 312 |
+
|
| 313 |
+
Here is the question. Answer it and don't confuse it with the previous conversation.
|
| 314 |
+
Question: {}
|
| 315 |
+
|
| 316 |
+
The answer is:
|
| 317 |
+
'''
|
| 318 |
+
|
| 319 |
+
class ESTP_contextualQ_benchmark:
|
| 320 |
+
def __init__(self, data, config=None):
|
| 321 |
+
self.data = data
|
| 322 |
+
|
| 323 |
+
task2number = {
|
| 324 |
+
"Object Relative Context": 0,
|
| 325 |
+
"Task Relative Context": 0,
|
| 326 |
+
}
|
| 327 |
+
|
| 328 |
+
for k,v in data.items():
|
| 329 |
+
for kk,vv in v.items():
|
| 330 |
+
for ll in vv:
|
| 331 |
+
task2number[ll['Task Type']] += 1
|
| 332 |
+
|
| 333 |
+
print_json(task2number)
|
| 334 |
+
|
| 335 |
+
self.config = config
|
| 336 |
+
|
| 337 |
+
def eval(self, data, model, output_path, eval_mode):
|
| 338 |
+
"""
|
| 339 |
+
Evaluate the model on the given data and update the data with the model responses.
|
| 340 |
+
data: data input
|
| 341 |
+
model: model to evaluate
|
| 342 |
+
"""
|
| 343 |
+
video_root = self.config.video_root
|
| 344 |
+
os.makedirs(os.path.dirname(output_path), exist_ok=True)
|
| 345 |
+
|
| 346 |
+
for k,v in tqdm.tqdm(data.items()):
|
| 347 |
+
for kk,vv in v.items():
|
| 348 |
+
for qa in vv:
|
| 349 |
+
if model.name() in qa.keys():
|
| 350 |
+
continue
|
| 351 |
+
video_path = os.path.join(video_root, k + '.mp4')
|
| 352 |
+
|
| 353 |
+
|
| 354 |
+
start_time = qa['clip_start_time'] if 'clip_start_time' in qa.keys() else qa['start_time']
|
| 355 |
+
end_time = qa['clip_end_time'] if 'clip_end_time' in qa.keys() else qa['end_time']
|
| 356 |
+
|
| 357 |
+
for conv in qa['conversation']:
|
| 358 |
+
if 'time' not in conv.keys():
|
| 359 |
+
conv['time'] = conv['start_time'] + (conv['end_time'] - conv['start_time']) / 2
|
| 360 |
+
qa['conversation'] = sorted(qa['conversation'], key=lambda x: x['time'])
|
| 361 |
+
|
| 362 |
+
max_time = qa['conversation'][-1]['end_time']
|
| 363 |
+
|
| 364 |
+
user_query, query_time = list_user_query(qa['conversation'])
|
| 365 |
+
|
| 366 |
+
start_time = min([start_time]+query_time)
|
| 367 |
+
end_time = min([end_time,max_time])
|
| 368 |
+
|
| 369 |
+
# if eval_mode == "frame_by_frame":
|
| 370 |
+
# dialog_history = eval_fbf_contextual(model, video_path, user_query, start_time, max_time, query_time)
|
| 371 |
+
# elif eval_mode == "passive_inference":
|
| 372 |
+
# dialog_history = eval_passive_inference_contextual(model, video_path, user_query, start_time, end_time, query_time)
|
| 373 |
+
# else:
|
| 374 |
+
# raise ValueError(f"Invalid eval mode: {eval_mode}")
|
| 375 |
+
|
| 376 |
+
try:
|
| 377 |
+
if eval_mode == "frame_by_frame":
|
| 378 |
+
dialog_history = eval_fbf_contextual(model, video_path, user_query, start_time, max_time, query_time)
|
| 379 |
+
elif eval_mode == "passive_inference":
|
| 380 |
+
dialog_history = eval_passive_inference_contextual(model, video_path, user_query, start_time, end_time, query_time)
|
| 381 |
+
else:
|
| 382 |
+
raise ValueError(f"Invalid eval mode: {eval_mode}")
|
| 383 |
+
except Exception as e:
|
| 384 |
+
print(e)
|
| 385 |
+
continue
|
| 386 |
+
|
| 387 |
+
qa[model.name()] = dialog_history
|
| 388 |
+
|
| 389 |
+
json.dump(data, open(output_path, 'w'), indent=4)
|
| 390 |
+
|
| 391 |
+
|
| 392 |
+
def dialog_history_to_context(dialog_history):
|
| 393 |
+
context = ""
|
| 394 |
+
for i in range(len(dialog_history)):
|
| 395 |
+
if dialog_history[i]['role'] == 'user':
|
| 396 |
+
context += f"At timestamp {dialog_history[i]['time']}, the following question occurred: {dialog_history[i]['content']} \n "
|
| 397 |
+
elif dialog_history[i]['role'] == 'assistant':
|
| 398 |
+
context += f"At timestamp {dialog_history[i]['time']}, the following answer occurred: {dialog_history[i]['content']} \n "
|
| 399 |
+
return context
|
| 400 |
+
|
| 401 |
+
def eval_fbf_contextual(model, video_path, questions, start_time, end_time, query_times=[]):
|
| 402 |
+
"""
|
| 403 |
+
Evaluate the model on the data frame by frame
|
| 404 |
+
|
| 405 |
+
Args:
|
| 406 |
+
model: The model to be evaluated
|
| 407 |
+
video_path: The path to the video file
|
| 408 |
+
inp: The input data
|
| 409 |
+
start_time: The start time of the evaluation
|
| 410 |
+
end_time: The end time of the evaluation
|
| 411 |
+
|
| 412 |
+
Returns:
|
| 413 |
+
None
|
| 414 |
+
"""
|
| 415 |
+
if model.name() in ONLINE_MODEL:
|
| 416 |
+
dialog_history = model.Run(video_path, questions, start_time, end_time, query_times)
|
| 417 |
+
elif model.name() in STREAMING_MODEL:
|
| 418 |
+
raise NotImplementedError(f"Streaming model {model.name()} is not implemented")
|
| 419 |
+
elif model.name() in GROUNDING_MODEL:
|
| 420 |
+
raise NotImplementedError(f"Grounding model {model.name()} is not implemented")
|
| 421 |
+
else:
|
| 422 |
+
assert isinstance(questions, list), "inp must be a list"
|
| 423 |
+
|
| 424 |
+
dialog_history = []
|
| 425 |
+
frame_fps = model.frame_fps
|
| 426 |
+
timecosts = []
|
| 427 |
+
context = ""
|
| 428 |
+
|
| 429 |
+
start_times = []
|
| 430 |
+
for question, query_time in zip(questions, query_times):
|
| 431 |
+
start_times.append(query_time)
|
| 432 |
+
end_times = start_times[1:] + [end_time]
|
| 433 |
+
|
| 434 |
+
for question, query_time, start_time, end_time in zip(questions, query_times, start_times, end_times):
|
| 435 |
+
# query = f"Is it the right time to answer the question \"{inp}\"? You need to answer yes or no first, and if yes, please answer the question."
|
| 436 |
+
query = f"Is it the right time to answer the question \"{question}\"? You need to answer yes or no."
|
| 437 |
+
first_inp = PROMPT_TEMPLATE_C_PROACTIVE.format(context, query)
|
| 438 |
+
|
| 439 |
+
yes_query = f"Please answer the question: \"{question}\""
|
| 440 |
+
yes_inp = PROMPT_TEMPLATE_C_PROACTIVE.format(context, yes_query)
|
| 441 |
+
|
| 442 |
+
current_time = min(start_time + 1 / frame_fps, end_time)
|
| 443 |
+
|
| 444 |
+
while current_time <= end_time:
|
| 445 |
+
real_start_time = time.time()
|
| 446 |
+
response, frame_number = model.Run(video_path, first_inp, start_time, current_time)
|
| 447 |
+
real_end_time = time.time()
|
| 448 |
+
timecosts.append(real_end_time - real_start_time)
|
| 449 |
+
|
| 450 |
+
|
| 451 |
+
if 'yes' in response.lower():
|
| 452 |
+
real_start_time = time.time()
|
| 453 |
+
response, frame_number = model.Run(video_path, yes_inp, start_time, current_time)
|
| 454 |
+
real_end_time = time.time()
|
| 455 |
+
timecosts.append(real_end_time - real_start_time)
|
| 456 |
+
|
| 457 |
+
fps = (current_time-start_time) * frame_fps / sum(timecosts)
|
| 458 |
+
|
| 459 |
+
dialog_history.append({
|
| 460 |
+
'role': 'user', 'content': yes_query, 'time': current_time, 'fps': fps, 'cost': timecosts[-1]
|
| 461 |
+
})
|
| 462 |
+
dialog_history.append({
|
| 463 |
+
'role': 'assistant', 'content': response, 'time': current_time, 'fps': fps, 'cost': timecosts[-1]
|
| 464 |
+
})
|
| 465 |
+
current_time += 1 / frame_fps
|
| 466 |
+
|
| 467 |
+
context = "Here are the contextual information related to the video. Please answer the questions based on the contextual information: "
|
| 468 |
+
context += dialog_history_to_context(dialog_history)
|
| 469 |
+
try:
|
| 470 |
+
dialog_history.append({
|
| 471 |
+
'role': 'fps', 'content': (end_times[-1] - start_times[0]) / sum(timecosts),
|
| 472 |
+
})
|
| 473 |
+
except:
|
| 474 |
+
print(f"Error in {model.name()}")
|
| 475 |
+
print(f"start_times: {start_times}")
|
| 476 |
+
print(f"end_times: {end_times}")
|
| 477 |
+
print(f"timecosts: {timecosts}")
|
| 478 |
+
print(f"video_path: {video_path}")
|
| 479 |
+
|
| 480 |
+
return dialog_history
|
| 481 |
+
|
| 482 |
+
def eval_streaming_contextual(model, video_path, questions, start_time, end_time, query_times=[]):
|
| 483 |
+
"""
|
| 484 |
+
Evaluate the model on the data streaming
|
| 485 |
+
"""
|
| 486 |
+
# TODO: implement this
|
| 487 |
+
dialog_history = model.Run(video_path, questions, start_time, end_time, query_times)
|
| 488 |
+
return dialog_history
|
| 489 |
+
|
| 490 |
+
|
| 491 |
+
def eval_passive_inference_contextual(model, video_path, questions, start_time, end_time, query_times=[]):
|
| 492 |
+
"""
|
| 493 |
+
Evaluate the model by first feeding all video frames and then getting response
|
| 494 |
+
|
| 495 |
+
Args:
|
| 496 |
+
model: The model to be evaluated
|
| 497 |
+
video_path: The path to the video file
|
| 498 |
+
inp: The input data
|
| 499 |
+
start_time: The start time of the evaluation
|
| 500 |
+
end_time: The end time of the evaluation
|
| 501 |
+
query_times: The times when the questions are asked
|
| 502 |
+
|
| 503 |
+
Returns:
|
| 504 |
+
None
|
| 505 |
+
"""
|
| 506 |
+
if model.name() in ONLINE_MODEL:
|
| 507 |
+
dialog_history = model.Run(video_path, questions, start_time, end_time, query_times)
|
| 508 |
+
elif model.name() in STREAMING_MODEL:
|
| 509 |
+
assert isinstance(questions, list), "inp must be a list"
|
| 510 |
+
|
| 511 |
+
# 1. travel each question and get response
|
| 512 |
+
dialog_history = []
|
| 513 |
+
frame_fps = model.frame_fps
|
| 514 |
+
|
| 515 |
+
start_times = []
|
| 516 |
+
for question, query_time in zip(questions, query_times):
|
| 517 |
+
start_times.append(query_time)
|
| 518 |
+
end_times = start_times[1:] + [end_time]
|
| 519 |
+
|
| 520 |
+
for question, query_time, start_time, end_time in zip(questions, query_times, start_times, end_times):
|
| 521 |
+
|
| 522 |
+
# 1.1 get response
|
| 523 |
+
conversation = model.Run(video_path, question, start_time, end_time, query_time)
|
| 524 |
+
# 3. generate dialog history
|
| 525 |
+
dialog_history.extend(conversation)
|
| 526 |
+
|
| 527 |
+
elif model.name() in GROUNDING_MODEL:
|
| 528 |
+
raise NotImplementedError(f"Grounding model {model.name()} is not implemented")
|
| 529 |
+
else:
|
| 530 |
+
assert isinstance(questions, list), "inp must be a list"
|
| 531 |
+
|
| 532 |
+
# 1. travel each question and get response
|
| 533 |
+
dialog_history = []
|
| 534 |
+
frame_fps = model.frame_fps
|
| 535 |
+
context = ""
|
| 536 |
+
|
| 537 |
+
start_times = []
|
| 538 |
+
for question, query_time in zip(questions, query_times):
|
| 539 |
+
start_times.append(query_time)
|
| 540 |
+
end_times = start_times[1:] + [end_time]
|
| 541 |
+
|
| 542 |
+
|
| 543 |
+
for question, query_time, start_time, end_time in zip(questions, query_times, start_times, end_times):
|
| 544 |
+
|
| 545 |
+
if end_time <= start_time:
|
| 546 |
+
breakpoint()
|
| 547 |
+
|
| 548 |
+
|
| 549 |
+
query = question
|
| 550 |
+
inp = PROMPT_TEMPLATE_C_PASSIVE.format(context,query)
|
| 551 |
+
|
| 552 |
+
real_start_time = time.time()
|
| 553 |
+
response, frame_number = model.Run(video_path, inp, start_time, end_time)
|
| 554 |
+
real_end_time = time.time()
|
| 555 |
+
|
| 556 |
+
frame_fps = frame_number / (end_time - start_time)
|
| 557 |
+
|
| 558 |
+
# 2. parse response
|
| 559 |
+
response_list = response.split('\n')
|
| 560 |
+
response_list = [i.strip() for i in response_list if i.strip()]
|
| 561 |
+
answer_pairs = []
|
| 562 |
+
|
| 563 |
+
for response in response_list:
|
| 564 |
+
if '[' in response and ']' in response:
|
| 565 |
+
frame_idx = response.split('[')[1].split(']')[0]
|
| 566 |
+
|
| 567 |
+
try:
|
| 568 |
+
if ',' in frame_idx:
|
| 569 |
+
frame_idxs = [int(x.strip()) for x in frame_idx.split(',')]
|
| 570 |
+
elif '-' in frame_idx:
|
| 571 |
+
# Handle range format like "1-7"
|
| 572 |
+
start, end = map(int, frame_idx.split('-'))
|
| 573 |
+
frame_idxs = [float((start+end) / 2)]
|
| 574 |
+
else:
|
| 575 |
+
frame_idxs = [float(frame_idx)]
|
| 576 |
+
answer = response.split('[')[1].split(']')[1].strip()
|
| 577 |
+
for frame_idx in frame_idxs:
|
| 578 |
+
answer_time = start_time + frame_idx / frame_fps
|
| 579 |
+
answer_pairs.append((answer_time, answer))
|
| 580 |
+
except:
|
| 581 |
+
continue
|
| 582 |
+
|
| 583 |
+
# 3. generate dialog history
|
| 584 |
+
dialog_history.append({
|
| 585 |
+
'role': 'user',
|
| 586 |
+
'content': question,
|
| 587 |
+
'time': query_time,
|
| 588 |
+
'fps': frame_number / (real_end_time - real_start_time),
|
| 589 |
+
'cost': real_end_time - real_start_time
|
| 590 |
+
})
|
| 591 |
+
|
| 592 |
+
for answer_time, answer in answer_pairs:
|
| 593 |
+
dialog_history.append({
|
| 594 |
+
'role': 'assistant',
|
| 595 |
+
'content': answer,
|
| 596 |
+
'time': answer_time,
|
| 597 |
+
'fps': frame_number / (real_end_time - real_start_time),
|
| 598 |
+
'cost': real_end_time - real_start_time
|
| 599 |
+
})
|
| 600 |
+
|
| 601 |
+
|
| 602 |
+
context = "Here are the contextual information related to the video. Please answer the questions based on the contextual information: "
|
| 603 |
+
context += dialog_history_to_context(dialog_history)
|
| 604 |
+
|
| 605 |
+
|
| 606 |
+
return dialog_history
|
| 607 |
+
|
ESTP-Bench/estp_dataset/benchmark/eval.py
ADDED
|
@@ -0,0 +1,428 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json,os
|
| 2 |
+
import numpy as np
|
| 3 |
+
|
| 4 |
+
def load_multiple_json(file_path):
|
| 5 |
+
"""读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。"""
|
| 6 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 7 |
+
content = f.read()
|
| 8 |
+
|
| 9 |
+
decoder = json.JSONDecoder()
|
| 10 |
+
pos = 0
|
| 11 |
+
results = []
|
| 12 |
+
content_length = len(content)
|
| 13 |
+
|
| 14 |
+
while pos < content_length:
|
| 15 |
+
# 跳过空白字符
|
| 16 |
+
while pos < content_length and content[pos].isspace():
|
| 17 |
+
pos += 1
|
| 18 |
+
if pos >= content_length:
|
| 19 |
+
break
|
| 20 |
+
try:
|
| 21 |
+
obj, new_pos = decoder.raw_decode(content, pos)
|
| 22 |
+
results.append(obj)
|
| 23 |
+
pos = new_pos
|
| 24 |
+
except json.JSONDecodeError as e:
|
| 25 |
+
# 出现解析错误则退出循环
|
| 26 |
+
print(f"JSON 解析错误: {e}")
|
| 27 |
+
break
|
| 28 |
+
return results
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
import argparse
|
| 32 |
+
|
| 33 |
+
def parse_args():
|
| 34 |
+
parser = argparse.ArgumentParser()
|
| 35 |
+
parser.add_argument('--eval_file', type=str, default='')
|
| 36 |
+
parser.add_argument('--eval_mode', type=str, default='all')
|
| 37 |
+
return parser.parse_args()
|
| 38 |
+
|
| 39 |
+
args = parse_args()
|
| 40 |
+
|
| 41 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_llama_5_5.json' # tmp_predict_VideollmOnline_v2_correctness
|
| 42 |
+
eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_deepseek_1_2.json'
|
| 43 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5casesevaluator_deepseek_5_5.json'
|
| 44 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_5_5.json'
|
| 45 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_1_2.json'
|
| 46 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2evaluator_deepseek_1_2.json'
|
| 47 |
+
eval_model = 'MiniCPMV'
|
| 48 |
+
|
| 49 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json'
|
| 50 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_1_2.json'
|
| 51 |
+
# # eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LIVE_IT0.95.json'
|
| 52 |
+
# eval_model = 'VideollmOnline' # VideollmOnline MiniCPMV
|
| 53 |
+
|
| 54 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json'
|
| 55 |
+
# eval_model = 'MMDuet'
|
| 56 |
+
|
| 57 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json'
|
| 58 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_fitVal_5_cases_v2evaluator_llama_5_5.json'
|
| 59 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3Firstevaluator_llama_5_5.json'
|
| 60 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'
|
| 61 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'
|
| 62 |
+
|
| 63 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.6evaluator_llama_5_5.json'
|
| 64 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.7evaluator_llama_5_5.json'
|
| 65 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3highv2evaluator_llama_5_5.json'
|
| 66 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3firstv2evaluator_llama_5_5.json'
|
| 67 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3HighRegion_evaluator_llama_5_5.json'
|
| 68 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/train_/estpBenchSq5Cases_fbf_beaconlivel_h_stage2_v2evaluator_deepseek_5_5.json'
|
| 69 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 70 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage2evaluator_deepseek_5_5.json'
|
| 71 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high0.31_11evaluator_deepseek_1_2.json'
|
| 72 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 73 |
+
|
| 74 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 75 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json'
|
| 76 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_1_lowevaluator_deepseek_1_2.json'
|
| 77 |
+
eval_model = 'EWO'
|
| 78 |
+
|
| 79 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json'
|
| 80 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 81 |
+
|
| 82 |
+
# eval_model = 'LLaVAOneVision'
|
| 83 |
+
|
| 84 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passiveevaluator_deepseek_1_2.json'
|
| 85 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 86 |
+
# eval_model = 'LLaVANextVideo7B'
|
| 87 |
+
|
| 88 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_5_5.json'
|
| 89 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 90 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_passiveevaluator_deepseek_1_2.json'
|
| 91 |
+
# eval_model = 'Qwen2VL'
|
| 92 |
+
|
| 93 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json'
|
| 94 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json'
|
| 95 |
+
# eval_model = 'InternVLV28'
|
| 96 |
+
|
| 97 |
+
|
| 98 |
+
# eval_model = 'Lavila'
|
| 99 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Lavila_streaming_v2evaluator_deepseek_1_2.json'
|
| 100 |
+
|
| 101 |
+
# eval_model = 'EgoVLP'
|
| 102 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json'
|
| 103 |
+
|
| 104 |
+
# eval_model = 'CLIP'
|
| 105 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json'
|
| 106 |
+
|
| 107 |
+
|
| 108 |
+
parent_dir = os.path.dirname(eval_file)
|
| 109 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file.split('/')[-1])]
|
| 110 |
+
eval_result = []
|
| 111 |
+
for eval_file in eval_files:
|
| 112 |
+
eval_result += load_multiple_json(eval_file)
|
| 113 |
+
|
| 114 |
+
|
| 115 |
+
task2number = {
|
| 116 |
+
"Object Recognition": 0,
|
| 117 |
+
"Attribute Perception": 0,
|
| 118 |
+
"Text-Rich Understanding": 0,
|
| 119 |
+
"Object Localization": 0,
|
| 120 |
+
"Object State Change Recognition": 0,
|
| 121 |
+
"Ego Object Localization": 0,
|
| 122 |
+
"Ego Object State Change Recognition": 0,
|
| 123 |
+
"Action Recognition": 0,
|
| 124 |
+
"Object Function": 0,
|
| 125 |
+
"Information Function": 0,
|
| 126 |
+
"Action Reasoning": 0,
|
| 127 |
+
"Task Understanding": 0,
|
| 128 |
+
|
| 129 |
+
}
|
| 130 |
+
|
| 131 |
+
task2score = {
|
| 132 |
+
"Object Recognition": 0,
|
| 133 |
+
"Attribute Perception": 0,
|
| 134 |
+
"Text-Rich Understanding": 0,
|
| 135 |
+
"Object Localization": 0,
|
| 136 |
+
"Object State Change Recognition": 0,
|
| 137 |
+
"Ego Object Localization": 0,
|
| 138 |
+
"Ego Object State Change Recognition": 0,
|
| 139 |
+
"Action Recognition": 0,
|
| 140 |
+
"Object Function": 0,
|
| 141 |
+
"Information Function": 0,
|
| 142 |
+
"Action Reasoning": 0,
|
| 143 |
+
"Task Understanding": 0,
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
}
|
| 147 |
+
|
| 148 |
+
task2recall = {
|
| 149 |
+
"Object Recognition": 0,
|
| 150 |
+
"Attribute Perception": 0,
|
| 151 |
+
"Text-Rich Understanding": 0,
|
| 152 |
+
"Object Localization": 0,
|
| 153 |
+
"Object State Change Recognition": 0,
|
| 154 |
+
"Ego Object Localization": 0,
|
| 155 |
+
"Ego Object State Change Recognition": 0,
|
| 156 |
+
"Action Recognition": 0,
|
| 157 |
+
"Object Function": 0,
|
| 158 |
+
"Information Function": 0,
|
| 159 |
+
"Action Reasoning": 0,
|
| 160 |
+
"Task Understanding": 0,
|
| 161 |
+
|
| 162 |
+
|
| 163 |
+
}
|
| 164 |
+
|
| 165 |
+
task2recall_score = {
|
| 166 |
+
"Object Recognition": 0,
|
| 167 |
+
"Attribute Perception": 0,
|
| 168 |
+
"Text-Rich Understanding": 0,
|
| 169 |
+
"Object Localization": 0,
|
| 170 |
+
"Object State Change Recognition": 0,
|
| 171 |
+
"Ego Object Localization": 0,
|
| 172 |
+
"Ego Object State Change Recognition": 0,
|
| 173 |
+
"Action Recognition": 0,
|
| 174 |
+
"Object Function": 0,
|
| 175 |
+
"Information Function": 0,
|
| 176 |
+
"Action Reasoning": 0,
|
| 177 |
+
"Task Understanding": 0,
|
| 178 |
+
|
| 179 |
+
|
| 180 |
+
}
|
| 181 |
+
|
| 182 |
+
task2nonrecall_pred = {
|
| 183 |
+
"Object Recognition": 0,
|
| 184 |
+
"Attribute Perception": 0,
|
| 185 |
+
"Text-Rich Understanding": 0,
|
| 186 |
+
"Object Localization": 0,
|
| 187 |
+
"Object State Change Recognition": 0,
|
| 188 |
+
"Ego Object Localization": 0,
|
| 189 |
+
"Ego Object State Change Recognition": 0,
|
| 190 |
+
"Action Recognition": 0,
|
| 191 |
+
"Object Function": 0,
|
| 192 |
+
"Information Function": 0,
|
| 193 |
+
"Action Reasoning": 0,
|
| 194 |
+
"Task Understanding": 0,
|
| 195 |
+
|
| 196 |
+
}
|
| 197 |
+
|
| 198 |
+
task2precision = {
|
| 199 |
+
"Object Recognition": 0,
|
| 200 |
+
"Attribute Perception": 0,
|
| 201 |
+
"Text-Rich Understanding": 0,
|
| 202 |
+
"Object Localization": 0,
|
| 203 |
+
"Object State Change Recognition": 0,
|
| 204 |
+
"Ego Object Localization": 0,
|
| 205 |
+
"Ego Object State Change Recognition": 0,
|
| 206 |
+
"Action Recognition": 0,
|
| 207 |
+
"Object Function": 0,
|
| 208 |
+
"Information Function": 0,
|
| 209 |
+
"Action Reasoning": 0,
|
| 210 |
+
"Task Understanding": 0,
|
| 211 |
+
}
|
| 212 |
+
|
| 213 |
+
def scoreMean(score):
|
| 214 |
+
score = score.max(axis=0)
|
| 215 |
+
return score.mean()
|
| 216 |
+
|
| 217 |
+
|
| 218 |
+
def validScoreMean(score):
|
| 219 |
+
# FP
|
| 220 |
+
nonrecall_pred = (score.max(axis=0) == 0).sum()
|
| 221 |
+
|
| 222 |
+
# valid_score
|
| 223 |
+
valid_score = np.zeros(score.shape[0])
|
| 224 |
+
valid_recall = np.zeros(score.shape[0])
|
| 225 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 226 |
+
for i,s in enumerate(score):
|
| 227 |
+
valid_s = s[s > 0]
|
| 228 |
+
if len(valid_s) > 0:
|
| 229 |
+
valid_score[i] = valid_s.sum() / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 230 |
+
valid_score[i] = valid_s.sum() / (valid_s.sum() + (nonrecall_pred / score.shape[0]))
|
| 231 |
+
# valid_score[i] = valid_s.max() / (1 + (nonrecall_pred / score.shape[0]))
|
| 232 |
+
# valid_score[i] = valid_s.max() * len(valid_s) / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 233 |
+
|
| 234 |
+
valid_recall[i] = 1
|
| 235 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 236 |
+
return valid_score.mean(), valid_recall.mean(), valid_recall_score.mean(), nonrecall_pred
|
| 237 |
+
|
| 238 |
+
BETA = 1
|
| 239 |
+
def validScoreF1(score):
|
| 240 |
+
|
| 241 |
+
if score.shape[1] == 0:
|
| 242 |
+
return 0, 0, 0, 0, 0
|
| 243 |
+
# FP
|
| 244 |
+
FP = (score.max(axis=0) == 0).sum()
|
| 245 |
+
# TP
|
| 246 |
+
TP = 0
|
| 247 |
+
|
| 248 |
+
# valid_score
|
| 249 |
+
valid_score = np.zeros(score.shape[0])
|
| 250 |
+
valid_recall = np.zeros(score.shape[0])
|
| 251 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 252 |
+
for i,s in enumerate(score):
|
| 253 |
+
valid_s = s[s > 0]
|
| 254 |
+
if len(valid_s) > 0:
|
| 255 |
+
# four type compute text-time precision
|
| 256 |
+
valid_score[i] = valid_s.sum() / len(valid_s)
|
| 257 |
+
# valid_score[i] = valid_s.max()
|
| 258 |
+
# valid_score[i] = valid_s.sum()
|
| 259 |
+
|
| 260 |
+
valid_recall[i] = 1
|
| 261 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 262 |
+
# valid_recall_score[i] = valid_s.max()
|
| 263 |
+
# valid_score[i] = valid_s.sum()
|
| 264 |
+
|
| 265 |
+
TP = valid_score.sum()
|
| 266 |
+
precision = TP / (TP + FP)
|
| 267 |
+
recall = valid_recall_score.mean()
|
| 268 |
+
# breakpoint()
|
| 269 |
+
if precision == 0 or recall == 0:
|
| 270 |
+
F1 = 0
|
| 271 |
+
else:
|
| 272 |
+
F1 = (1 + BETA**2) * precision * recall / ((BETA**2 * precision) + recall)
|
| 273 |
+
|
| 274 |
+
F1 = 2*TP / (2*TP + FP + score.shape[0] - valid_recall.sum())
|
| 275 |
+
return F1, valid_recall.mean(), valid_recall_score.mean(), FP, precision
|
| 276 |
+
|
| 277 |
+
def topkValidScoreMean(score, k=10):
|
| 278 |
+
score = score.max(axis=1)
|
| 279 |
+
score = score[:k]
|
| 280 |
+
valid_score = score[score > 0]
|
| 281 |
+
return valid_score.mean()
|
| 282 |
+
|
| 283 |
+
total_fps = 0
|
| 284 |
+
total_kv_cache_size = 0
|
| 285 |
+
total_response_number = 0
|
| 286 |
+
total_answer_number = 0
|
| 287 |
+
total_precision = 0
|
| 288 |
+
for ll in eval_result:
|
| 289 |
+
if eval_model in ll.keys():
|
| 290 |
+
total_answer_number += len(ll['conversation'])
|
| 291 |
+
this_turn_response_number = 0
|
| 292 |
+
for response in ll[eval_model]:
|
| 293 |
+
if response['role'].lower() == 'assistant':
|
| 294 |
+
if 'fps' in response:
|
| 295 |
+
total_fps+=response['fps']
|
| 296 |
+
total_response_number += 1
|
| 297 |
+
this_turn_response_number += 1
|
| 298 |
+
if 'kv_cache_size' in response:
|
| 299 |
+
total_kv_cache_size += response['kv_cache_size']
|
| 300 |
+
task2number[ll['Task Type'].strip()] += 1
|
| 301 |
+
# breakpoint()
|
| 302 |
+
text_score = np.array(ll['evaluator_output_text']) / 10
|
| 303 |
+
reponse_score = np.array(ll['evaluator_output_reponse']) / 10
|
| 304 |
+
# if ll['Task Type'] == 'Ego Object State Change Recognition':
|
| 305 |
+
if args.eval_mode == 'all' and eval_model not in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 306 |
+
score = (text_score+reponse_score)
|
| 307 |
+
elif args.eval_mode == 'text':
|
| 308 |
+
score = text_score
|
| 309 |
+
elif args.eval_mode == 'response' or eval_model in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 310 |
+
score = reponse_score
|
| 311 |
+
score_mean, recall_mean, recall_score_mean, nonrecall_pred, precision = validScoreF1(score)
|
| 312 |
+
task2score[ll['Task Type'].strip()] += score_mean
|
| 313 |
+
task2recall[ll['Task Type'].strip()] += recall_mean
|
| 314 |
+
task2recall_score[ll['Task Type'].strip()] += recall_score_mean
|
| 315 |
+
task2nonrecall_pred[ll['Task Type'].strip()] += nonrecall_pred
|
| 316 |
+
task2precision[ll['Task Type'].strip()] += precision
|
| 317 |
+
total_precision += nonrecall_pred / this_turn_response_number if this_turn_response_number > 0 else 0
|
| 318 |
+
|
| 319 |
+
|
| 320 |
+
|
| 321 |
+
for k,v in task2score.items():
|
| 322 |
+
if task2number[k] == 0:
|
| 323 |
+
task2score[k] = 0
|
| 324 |
+
task2recall[k] = 0
|
| 325 |
+
task2recall_score[k] = 0
|
| 326 |
+
task2nonrecall_pred[k] = 0
|
| 327 |
+
task2precision[k] = 0
|
| 328 |
+
else:
|
| 329 |
+
task2score[k] = v / task2number[k] * 100
|
| 330 |
+
task2recall[k] = task2recall[k] / task2number[k] * 100
|
| 331 |
+
task2recall_score[k] = task2recall_score[k] / task2number[k] * 100
|
| 332 |
+
task2nonrecall_pred[k] = task2nonrecall_pred[k] / task2number[k]
|
| 333 |
+
task2precision[k] = task2precision[k] / task2number[k] * 100
|
| 334 |
+
|
| 335 |
+
print(json.dumps(task2number, indent=4))
|
| 336 |
+
print(json.dumps({k: round(v, 2) for k,v in task2score.items()}, indent=4))
|
| 337 |
+
print(json.dumps(task2recall, indent=4))
|
| 338 |
+
# print(json.dumps(task2recall_score, indent=4))
|
| 339 |
+
# print(json.dumps(task2nonrecall_pred, indent=4))
|
| 340 |
+
|
| 341 |
+
print("Total question number: ", sum(task2number.values()))
|
| 342 |
+
|
| 343 |
+
print(f"Average Score: {sum(task2score.values())/len(task2number.values())}")
|
| 344 |
+
|
| 345 |
+
# Calculate average for first 8 task types
|
| 346 |
+
first_8_tasks = list(task2number.keys())[:8]
|
| 347 |
+
first_8_scores = [task2score[task] for task in first_8_tasks]
|
| 348 |
+
first_8_valid_scores = [score for score in first_8_scores if score > 0]
|
| 349 |
+
first_8_recall_scores = [task2recall[task] for task in first_8_tasks]
|
| 350 |
+
try:
|
| 351 |
+
print(f"Average Score (First 8 Tasks): {sum(first_8_valid_scores)/len(first_8_valid_scores):.2f}")
|
| 352 |
+
except:
|
| 353 |
+
print(f"Average Score (First 8 Tasks): 0.0")
|
| 354 |
+
|
| 355 |
+
# Calculate average for last 4 task types
|
| 356 |
+
last_4_tasks = list(task2number.keys())[-4:]
|
| 357 |
+
last_4_scores = [task2score[task] for task in last_4_tasks]
|
| 358 |
+
last_4_valid_scores = [score for score in last_4_scores if score > 0]
|
| 359 |
+
last_4_recall_scores = [task2recall[task] for task in last_4_tasks]
|
| 360 |
+
try:
|
| 361 |
+
print(f"Average Score (Last 4 Tasks): {sum(last_4_valid_scores)/len(last_4_valid_scores):.2f}")
|
| 362 |
+
except:
|
| 363 |
+
print(f"Average Score (Last 4 Tasks): 0.0")
|
| 364 |
+
|
| 365 |
+
|
| 366 |
+
print(f"Average Recall: {sum(task2recall.values())/len(task2number.values())}")
|
| 367 |
+
print(f"Average Recall Score: {sum(task2recall_score.values())/len(task2number.values())}")
|
| 368 |
+
print(f"Average Precision: {sum(task2precision.values())/len(task2number.values())}")
|
| 369 |
+
|
| 370 |
+
|
| 371 |
+
print(f"Total Question Number: {sum(task2number.values())}")
|
| 372 |
+
print(f"Average FPS: {total_fps/total_response_number}")
|
| 373 |
+
print(f"Average KV Cache: {total_kv_cache_size/total_response_number}")
|
| 374 |
+
print(f"Average Non-Recall Pred: {sum(task2nonrecall_pred.values())/sum(task2number.values())}")
|
| 375 |
+
print(f"Average Response Number: {total_response_number/sum(task2number.values())}")
|
| 376 |
+
print(f"Average Precision: {total_precision/sum(task2number.values()) * 100}")
|
| 377 |
+
|
| 378 |
+
|
| 379 |
+
print(f"[{sum(task2recall.values())/len(task2number.values()):.1f}, {sum(task2precision.values())/len(task2number.values()):.1f}]")
|
| 380 |
+
|
| 381 |
+
def generate_latex_table_row():
|
| 382 |
+
# Generate LaTeX format table row
|
| 383 |
+
print("\n# LaTeX format table row")
|
| 384 |
+
print(f"& {eval_model} ", end="")
|
| 385 |
+
|
| 386 |
+
# First 8 tasks
|
| 387 |
+
for task in list(task2number.keys())[:8]:
|
| 388 |
+
print(f"& {task2score[task]:.1f} ", end="")
|
| 389 |
+
|
| 390 |
+
# Average of first 8 tasks
|
| 391 |
+
print(f"& {sum(first_8_valid_scores)/len(first_8_valid_scores):.1f} ", end="")
|
| 392 |
+
|
| 393 |
+
# Last 4 tasks
|
| 394 |
+
for task in list(task2number.keys())[-4:]:
|
| 395 |
+
print(f"& {task2score[task]:.1f} ", end="")
|
| 396 |
+
|
| 397 |
+
# Average of last 4 tasks
|
| 398 |
+
print(f"& {sum(last_4_valid_scores)/len(last_4_valid_scores):.1f} ", end="")
|
| 399 |
+
|
| 400 |
+
# Overall average
|
| 401 |
+
# print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")
|
| 402 |
+
|
| 403 |
+
generate_latex_table_row()
|
| 404 |
+
|
| 405 |
+
|
| 406 |
+
# def generate_latex_table_row():
|
| 407 |
+
# # Generate LaTeX format table row
|
| 408 |
+
# print("\n# LaTeX format table row")
|
| 409 |
+
# print(f"& {eval_model} ", end="")
|
| 410 |
+
|
| 411 |
+
# # First 8 tasks
|
| 412 |
+
# for task in list(task2number.keys())[:8]:
|
| 413 |
+
# print(f"& {task2recall[task]:.1f} ", end="")
|
| 414 |
+
|
| 415 |
+
# # Average of first 8 tasks
|
| 416 |
+
# print(f"& {sum(first_8_recall_scores)/len(first_8_recall_scores):.1f} ", end="")
|
| 417 |
+
|
| 418 |
+
# # Last 4 tasks
|
| 419 |
+
# for task in list(task2number.keys())[-4:]:
|
| 420 |
+
# print(f"& {task2recall[task]:.1f} ", end="")
|
| 421 |
+
|
| 422 |
+
# # Average of last 4 tasks
|
| 423 |
+
# print(f"& {sum(last_4_recall_scores)/len(last_4_recall_scores):.1f} ", end="")
|
| 424 |
+
|
| 425 |
+
# # Overall average
|
| 426 |
+
# # print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")
|
| 427 |
+
|
| 428 |
+
# generate_latex_table_row()
|
ESTP-Bench/estp_dataset/benchmark/eval_cost.py
ADDED
|
@@ -0,0 +1,170 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json,os
|
| 2 |
+
import numpy as np
|
| 3 |
+
|
| 4 |
+
def load_multiple_json(file_path):
|
| 5 |
+
"""读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。"""
|
| 6 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 7 |
+
content = f.read()
|
| 8 |
+
|
| 9 |
+
decoder = json.JSONDecoder()
|
| 10 |
+
pos = 0
|
| 11 |
+
results = []
|
| 12 |
+
content_length = len(content)
|
| 13 |
+
|
| 14 |
+
while pos < content_length:
|
| 15 |
+
# 跳过空白字符
|
| 16 |
+
while pos < content_length and content[pos].isspace():
|
| 17 |
+
pos += 1
|
| 18 |
+
if pos >= content_length:
|
| 19 |
+
break
|
| 20 |
+
try:
|
| 21 |
+
obj, new_pos = decoder.raw_decode(content, pos)
|
| 22 |
+
results.append(obj)
|
| 23 |
+
pos = new_pos
|
| 24 |
+
except json.JSONDecodeError as e:
|
| 25 |
+
# 出现解析错误则退出循环
|
| 26 |
+
print(f"JSON 解析错误: {e}")
|
| 27 |
+
break
|
| 28 |
+
return results
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LIVE_IT0.95.json'
|
| 33 |
+
eval_model = 'VideollmOnline'
|
| 34 |
+
|
| 35 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage2.json'
|
| 36 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high11_.json'
|
| 37 |
+
# eval_model = 'EWO'
|
| 38 |
+
|
| 39 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175.json'
|
| 40 |
+
# eval_model = 'MiniCPMV'
|
| 41 |
+
|
| 42 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_5_5.json'
|
| 43 |
+
# eval_model = 'LLaVAOneVision'
|
| 44 |
+
|
| 45 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passiveevaluator_deepseek_5_5.json'
|
| 46 |
+
# eval_model = 'LLaVANextVideo7B'
|
| 47 |
+
|
| 48 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_5_5.json'
|
| 49 |
+
# eval_model = 'Qwen2VL'
|
| 50 |
+
|
| 51 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_5_5.json'
|
| 52 |
+
# eval_model = 'InternVLV28'
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
parent_dir = os.path.dirname(eval_file)
|
| 56 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file.split('/')[-1])]
|
| 57 |
+
eval_result = {}
|
| 58 |
+
for eval_file in eval_files:
|
| 59 |
+
eval_result.update(json.load(open(eval_file)))
|
| 60 |
+
|
| 61 |
+
task2number = {
|
| 62 |
+
"Object Recognition": 0,
|
| 63 |
+
"Attribute Perception": 0,
|
| 64 |
+
"Text-Rich Understanding": 0,
|
| 65 |
+
"Object Localization": 0,
|
| 66 |
+
"Object State Change Recognition": 0,
|
| 67 |
+
"Ego Object Localization": 0,
|
| 68 |
+
"Ego Object State Change Recognition": 0,
|
| 69 |
+
"Action Recognition": 0,
|
| 70 |
+
"Object Function": 0,
|
| 71 |
+
"Information Function": 0,
|
| 72 |
+
"Action Reasoning": 0,
|
| 73 |
+
"Task Understanding": 0,
|
| 74 |
+
|
| 75 |
+
}
|
| 76 |
+
|
| 77 |
+
task2score = {
|
| 78 |
+
"Object Recognition": 0,
|
| 79 |
+
"Attribute Perception": 0,
|
| 80 |
+
"Text-Rich Understanding": 0,
|
| 81 |
+
"Object Localization": 0,
|
| 82 |
+
"Object State Change Recognition": 0,
|
| 83 |
+
"Ego Object Localization": 0,
|
| 84 |
+
"Ego Object State Change Recognition": 0,
|
| 85 |
+
"Action Recognition": 0,
|
| 86 |
+
"Object Function": 0,
|
| 87 |
+
"Information Function": 0,
|
| 88 |
+
"Action Reasoning": 0,
|
| 89 |
+
"Task Understanding": 0,
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
}
|
| 93 |
+
|
| 94 |
+
task2recall = {
|
| 95 |
+
"Object Recognition": 0,
|
| 96 |
+
"Attribute Perception": 0,
|
| 97 |
+
"Text-Rich Understanding": 0,
|
| 98 |
+
"Object Localization": 0,
|
| 99 |
+
"Object State Change Recognition": 0,
|
| 100 |
+
"Ego Object Localization": 0,
|
| 101 |
+
"Ego Object State Change Recognition": 0,
|
| 102 |
+
"Action Recognition": 0,
|
| 103 |
+
"Object Function": 0,
|
| 104 |
+
"Information Function": 0,
|
| 105 |
+
"Action Reasoning": 0,
|
| 106 |
+
"Task Understanding": 0,
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
}
|
| 110 |
+
|
| 111 |
+
task2recall_score = {
|
| 112 |
+
"Object Recognition": 0,
|
| 113 |
+
"Attribute Perception": 0,
|
| 114 |
+
"Text-Rich Understanding": 0,
|
| 115 |
+
"Object Localization": 0,
|
| 116 |
+
"Object State Change Recognition": 0,
|
| 117 |
+
"Ego Object Localization": 0,
|
| 118 |
+
"Ego Object State Change Recognition": 0,
|
| 119 |
+
"Action Recognition": 0,
|
| 120 |
+
"Object Function": 0,
|
| 121 |
+
"Information Function": 0,
|
| 122 |
+
"Action Reasoning": 0,
|
| 123 |
+
"Task Understanding": 0,
|
| 124 |
+
|
| 125 |
+
|
| 126 |
+
}
|
| 127 |
+
|
| 128 |
+
task2nonrecall_pred = {
|
| 129 |
+
"Object Recognition": 0,
|
| 130 |
+
"Attribute Perception": 0,
|
| 131 |
+
"Text-Rich Understanding": 0,
|
| 132 |
+
"Object Localization": 0,
|
| 133 |
+
"Object State Change Recognition": 0,
|
| 134 |
+
"Ego Object Localization": 0,
|
| 135 |
+
"Ego Object State Change Recognition": 0,
|
| 136 |
+
"Action Recognition": 0,
|
| 137 |
+
"Object Function": 0,
|
| 138 |
+
"Information Function": 0,
|
| 139 |
+
"Action Reasoning": 0,
|
| 140 |
+
"Task Understanding": 0,
|
| 141 |
+
|
| 142 |
+
}
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
total_fps_last = 0
|
| 148 |
+
total_fps = 0
|
| 149 |
+
total_response_number = 0
|
| 150 |
+
total_kv_cache_size = 0
|
| 151 |
+
for k,v in eval_result.items():
|
| 152 |
+
for kk,vv in v.items():
|
| 153 |
+
for ll in vv:
|
| 154 |
+
if eval_model in ll.keys():
|
| 155 |
+
for response in ll[eval_model]:
|
| 156 |
+
if response['role'] == 'fps':
|
| 157 |
+
total_fps_last+=response['content']
|
| 158 |
+
if response['role'].lower() == 'assistant':
|
| 159 |
+
if 'fps' in response:
|
| 160 |
+
total_fps+=response['fps']
|
| 161 |
+
total_response_number += 1
|
| 162 |
+
if 'kv_cache_size' in response:
|
| 163 |
+
total_kv_cache_size+=response['kv_cache_size']
|
| 164 |
+
task2number[ll['Task Type'].strip()] += 1
|
| 165 |
+
|
| 166 |
+
print('total_qa: ', sum(task2number.values()))
|
| 167 |
+
print(f"Average FPS: {total_fps/total_response_number}")
|
| 168 |
+
print(f"Average KV Cache: {total_kv_cache_size/total_response_number}")
|
| 169 |
+
print("total_fps_last_mean: ", total_fps_last / sum(task2number.values()))
|
| 170 |
+
print(f"Average Response Number: {total_response_number/sum(task2number.values())}")
|
ESTP-Bench/estp_dataset/benchmark/eval_cqa.py
ADDED
|
@@ -0,0 +1,347 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json,os
|
| 2 |
+
import numpy as np
|
| 3 |
+
|
| 4 |
+
def load_multiple_json(file_path):
|
| 5 |
+
"""读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。"""
|
| 6 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 7 |
+
content = f.read()
|
| 8 |
+
|
| 9 |
+
decoder = json.JSONDecoder()
|
| 10 |
+
pos = 0
|
| 11 |
+
results = []
|
| 12 |
+
content_length = len(content)
|
| 13 |
+
|
| 14 |
+
while pos < content_length:
|
| 15 |
+
# 跳过空白字符
|
| 16 |
+
while pos < content_length and content[pos].isspace():
|
| 17 |
+
pos += 1
|
| 18 |
+
if pos >= content_length:
|
| 19 |
+
break
|
| 20 |
+
try:
|
| 21 |
+
obj, new_pos = decoder.raw_decode(content, pos)
|
| 22 |
+
results.append(obj)
|
| 23 |
+
pos = new_pos
|
| 24 |
+
except json.JSONDecodeError as e:
|
| 25 |
+
# 出现解析错误则退出循环
|
| 26 |
+
print(f"JSON 解析错误: {e}")
|
| 27 |
+
break
|
| 28 |
+
return results
|
| 29 |
+
|
| 30 |
+
# import argparse
|
| 31 |
+
|
| 32 |
+
# def parse_args():
|
| 33 |
+
# parser = argparse.ArgumentParser()
|
| 34 |
+
# parser.add_argument('--eval_model', type=str, default='InternVLV28', help='Model to evaluate')
|
| 35 |
+
# parser.add_argument('--inference_mode', type=str, default='default', help='Evaluation mode: passive, fbf, streaming')
|
| 36 |
+
# return parser.parse_args()
|
| 37 |
+
|
| 38 |
+
# args = parse_args()
|
| 39 |
+
|
| 40 |
+
# # 使用命令行参数或默认值
|
| 41 |
+
# eval_model = args.eval_model
|
| 42 |
+
|
| 43 |
+
# # 根据选择的模型设置评估文件路径
|
| 44 |
+
# model_file_mapping = {
|
| 45 |
+
# 'MiniCPMV': {
|
| 46 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_passive_v2evaluator_deepseek_1_2.json',
|
| 47 |
+
# 'fbf': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175evaluator_deepseek_1_2.json'
|
| 48 |
+
# },
|
| 49 |
+
# 'Qwen2VL': {
|
| 50 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/Qwen2VL_passive_v2evaluator_deepseek_1_2.json',
|
| 51 |
+
# 'fbf': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175evaluator_deepseek_1_2.json'
|
| 52 |
+
# },
|
| 53 |
+
# 'LLaVAOneVision': {
|
| 54 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVAOneVision_passive_v2evaluator_deepseek_1_2.json',
|
| 55 |
+
# 'fbf': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_0.175evaluator_deepseek_1_2.json'
|
| 56 |
+
# },
|
| 57 |
+
# 'Lavila': {
|
| 58 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/Lavila_streaming_v2evaluator_deepseek_1_2.json'
|
| 59 |
+
# },
|
| 60 |
+
# 'EgoVLP': {
|
| 61 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json'
|
| 62 |
+
# },
|
| 63 |
+
# 'CLIP': {
|
| 64 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json'
|
| 65 |
+
# },
|
| 66 |
+
# 'MMDuet': {
|
| 67 |
+
# 'default': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/MMDuetevaluator_deepseek_1_2.json'
|
| 68 |
+
# },
|
| 69 |
+
# 'LLaVANextVideo7B': {
|
| 70 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVANextVideo7B_passive_v2evaluator_deepseek_1_2.json',
|
| 71 |
+
# 'fbf': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175evaluator_deepseek_1_2.json'
|
| 72 |
+
# },
|
| 73 |
+
# 'InternVLV28': {
|
| 74 |
+
# 'passive': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/InternVLV28_passive_v2evaluator_deepseek_1_2.json',
|
| 75 |
+
# 'fbf': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json'
|
| 76 |
+
# },
|
| 77 |
+
# 'VideollmOnline': {
|
| 78 |
+
# 'default': '/root/videollm-online/data/estp_dataset/estpCqa_baseline/VideollmOnline0.8evaluator_deepseek_1_2.json'
|
| 79 |
+
# }
|
| 80 |
+
# }
|
| 81 |
+
|
| 82 |
+
# # 默认使用fbf版本，如果没有则使用passive，如果都没有则使用默认或streaming
|
| 83 |
+
# if eval_model in model_file_mapping:
|
| 84 |
+
# # 展示可用的inference_mode选项
|
| 85 |
+
# available_modes = list(model_file_mapping[eval_model].keys())
|
| 86 |
+
# print(f"可用的inference_mode选项:")
|
| 87 |
+
# for i, mode in enumerate(available_modes, 1):
|
| 88 |
+
# print(f"{i}. {mode}")
|
| 89 |
+
|
| 90 |
+
# # 如果用户提供的inference_mode不在可用选项中，让用户选择
|
| 91 |
+
# if args.inference_mode not in available_modes:
|
| 92 |
+
# choice = input(f"请选择inference_mode (1-{len(available_modes)}): ")
|
| 93 |
+
# try:
|
| 94 |
+
# choice_idx = int(choice) - 1
|
| 95 |
+
# if 0 <= choice_idx < len(available_modes):
|
| 96 |
+
# args.inference_mode = available_modes[choice_idx]
|
| 97 |
+
# else:
|
| 98 |
+
# args.inference_mode = available_modes[0] # 默认使用第一个选项
|
| 99 |
+
# print(f"无效选择，使用默认模式: {args.inference_mode}")
|
| 100 |
+
# except ValueError:
|
| 101 |
+
# args.inference_mode = available_modes[0] # 默认使用第一个选项
|
| 102 |
+
# print(f"无效输入，使用默认模式: {args.inference_mode}")
|
| 103 |
+
|
| 104 |
+
# eval_file = model_file_mapping[eval_model][args.inference_mode]
|
| 105 |
+
# else:
|
| 106 |
+
# raise ValueError(f"未找到模型 {eval_model} 的评估文件")
|
| 107 |
+
|
| 108 |
+
# print(f"评估模型: {eval_model}")
|
| 109 |
+
# print(f"评估文件: {eval_file}")
|
| 110 |
+
|
| 111 |
+
|
| 112 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_passiveevaluator_deepseek_1_2.json' # tmp_predict_VideollmOnline_v2_correctness
|
| 113 |
+
eval_model = 'MiniCPMV'
|
| 114 |
+
|
| 115 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpCqa_baseline/Qwen2VL_passiveevaluator_deepseek_1_2.json' # tmp_predict_VideollmOnline_v2_correctness
|
| 116 |
+
eval_model = 'Qwen2VL'
|
| 117 |
+
|
| 118 |
+
eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_baseline/VideollmOnline0.9evaluator_deepseek_1_2.json'
|
| 119 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LIVE_IT0.95evaluator_deepseek_1_2.json'
|
| 120 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LIVE_IT_smoothing_v2evaluator_deepseek_1_2.json'
|
| 121 |
+
eval_model = 'VideollmOnline' # VideollmOnline MiniCPMV
|
| 122 |
+
|
| 123 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json' # tmp_predict_VideollmOnline_v2_correctness
|
| 124 |
+
# eval_model = 'LLaVAOneVision'
|
| 125 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage2_v3evaluator_deepseek_1_2.json'
|
| 126 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage2.5evaluator_deepseek_1_2.json'
|
| 127 |
+
eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage3.5_high0.31_11evaluator_deepseek_1_2.json'
|
| 128 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage3_v3evaluator_deepseek_1_2.json'
|
| 129 |
+
# eval_file = '/2022233235/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage3.5_v3evaluator_deepseek_1_2.json'
|
| 130 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage2_lowevaluator_deepseek_1_2.json'
|
| 131 |
+
eval_model = 'EWO'
|
| 132 |
+
|
| 133 |
+
parent_dir = os.path.dirname(eval_file)
|
| 134 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file.split('/')[-1])]
|
| 135 |
+
eval_result = []
|
| 136 |
+
for eval_file in eval_files:
|
| 137 |
+
eval_result += load_multiple_json(eval_file)
|
| 138 |
+
|
| 139 |
+
|
| 140 |
+
task2number = {
|
| 141 |
+
"Object Relative Context": 0,
|
| 142 |
+
"Task Relative Context": 0,
|
| 143 |
+
}
|
| 144 |
+
|
| 145 |
+
task2recall = {
|
| 146 |
+
"Object Relative Context": 0,
|
| 147 |
+
"Task Relative Context": 0,
|
| 148 |
+
}
|
| 149 |
+
|
| 150 |
+
task2recall_score = {
|
| 151 |
+
"Object Relative Context": 0,
|
| 152 |
+
"Task Relative Context": 0,
|
| 153 |
+
}
|
| 154 |
+
|
| 155 |
+
task2score = {
|
| 156 |
+
"Object Relative Context": 0,
|
| 157 |
+
"Task Relative Context": 0,
|
| 158 |
+
}
|
| 159 |
+
|
| 160 |
+
task2nonrecall_pred = {
|
| 161 |
+
"Object Relative Context": 0,
|
| 162 |
+
"Task Relative Context": 0,
|
| 163 |
+
}
|
| 164 |
+
|
| 165 |
+
task2precision = {
|
| 166 |
+
"Object Relative Context": 0,
|
| 167 |
+
"Task Relative Context": 0,
|
| 168 |
+
}
|
| 169 |
+
|
| 170 |
+
task2bug = {
|
| 171 |
+
"Object Relative Context": 0,
|
| 172 |
+
"Task Relative Context": 0,
|
| 173 |
+
}
|
| 174 |
+
|
| 175 |
+
def scoreMean(score):
|
| 176 |
+
score = score.max(axis=0)
|
| 177 |
+
return score.mean()
|
| 178 |
+
|
| 179 |
+
|
| 180 |
+
def validScoreMean(score):
|
| 181 |
+
# FP
|
| 182 |
+
if len(score) == 0:
|
| 183 |
+
return 0, 0, 0, 0
|
| 184 |
+
nonrecall_pred = (score.max(axis=0) == 0).sum()
|
| 185 |
+
|
| 186 |
+
# valid_score
|
| 187 |
+
valid_score = np.zeros(score.shape[0])
|
| 188 |
+
valid_recall = np.zeros(score.shape[0])
|
| 189 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 190 |
+
for i,s in enumerate(score):
|
| 191 |
+
valid_s = s[s > 0]
|
| 192 |
+
if len(valid_s) > 0:
|
| 193 |
+
valid_score[i] = valid_s.sum() / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 194 |
+
valid_score[i] = valid_s.sum() / (valid_s.sum() + (nonrecall_pred / score.shape[0]))
|
| 195 |
+
# valid_score[i] = valid_s.max() / (1 + (nonrecall_pred / score.shape[0]))
|
| 196 |
+
# valid_score[i] = valid_s.max() * len(valid_s) / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 197 |
+
|
| 198 |
+
valid_recall[i] = 1
|
| 199 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 200 |
+
return valid_score.mean(), valid_recall.mean(), valid_recall_score.mean(), nonrecall_pred
|
| 201 |
+
|
| 202 |
+
BETA = 1
|
| 203 |
+
def validScoreF1(score):
|
| 204 |
+
|
| 205 |
+
if len(score) == 0:
|
| 206 |
+
return 0, 0, 0, 0, 0
|
| 207 |
+
|
| 208 |
+
# FP
|
| 209 |
+
FP = (score.max(axis=0) == 0).sum()
|
| 210 |
+
# TP
|
| 211 |
+
TP = 0
|
| 212 |
+
|
| 213 |
+
# valid_score
|
| 214 |
+
valid_score = np.zeros(score.shape[0])
|
| 215 |
+
valid_recall = np.zeros(score.shape[0])
|
| 216 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 217 |
+
for i,s in enumerate(score):
|
| 218 |
+
valid_s = s[s > 0]
|
| 219 |
+
if len(valid_s) > 0:
|
| 220 |
+
# four type compute text-time precision
|
| 221 |
+
valid_score[i] = valid_s.sum() / len(valid_s)
|
| 222 |
+
# valid_score[i] = valid_s.max()
|
| 223 |
+
|
| 224 |
+
valid_recall[i] = 1
|
| 225 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 226 |
+
# valid_recall_score[i] = valid_s.max()
|
| 227 |
+
|
| 228 |
+
TP = valid_score.sum()
|
| 229 |
+
precision = TP / (TP + FP)
|
| 230 |
+
if np.isnan(precision):
|
| 231 |
+
precision = 0
|
| 232 |
+
recall = valid_recall_score.mean()
|
| 233 |
+
if precision == 0 or recall == 0:
|
| 234 |
+
F1 = 0
|
| 235 |
+
else:
|
| 236 |
+
F1 = (1 + BETA**2) * precision * recall / ((BETA**2 * precision) + recall)
|
| 237 |
+
|
| 238 |
+
# breakpoint()
|
| 239 |
+
F1 = 2*TP / (2*TP + FP + score.shape[0] - valid_recall.sum())
|
| 240 |
+
|
| 241 |
+
return F1, valid_recall.mean(), valid_recall_score.mean(), FP, precision
|
| 242 |
+
|
| 243 |
+
def topkValidScoreMean(score, k=10):
|
| 244 |
+
score = score.max(axis=1)
|
| 245 |
+
score = score[:k]
|
| 246 |
+
valid_score = score[score > 0]
|
| 247 |
+
return valid_score.mean()
|
| 248 |
+
|
| 249 |
+
total_fps = 0
|
| 250 |
+
total_kv_cache_size = 0
|
| 251 |
+
total_response_number = 0
|
| 252 |
+
total_answer_number = 0
|
| 253 |
+
for ll in eval_result:
|
| 254 |
+
if eval_model in ll.keys():
|
| 255 |
+
total_answer_number += len(ll['conversation'])
|
| 256 |
+
for response in ll[eval_model]:
|
| 257 |
+
if response['role'].lower() == 'assistant':
|
| 258 |
+
if 'fps' in response:
|
| 259 |
+
total_fps+=response['fps']
|
| 260 |
+
total_response_number += 1
|
| 261 |
+
if 'kv_cache_size' in response:
|
| 262 |
+
total_kv_cache_size += response['kv_cache_size']
|
| 263 |
+
task2number[ll['Task Type'].strip()] += 1
|
| 264 |
+
# breakpoint()
|
| 265 |
+
text_score = np.array(ll['evaluator_output_text']) / 10
|
| 266 |
+
reponse_score = np.array(ll['evaluator_output_reponse']) / 10
|
| 267 |
+
# if ll['Task Type'] == 'Ego Object State Change Recognition':
|
| 268 |
+
score = (text_score+reponse_score)
|
| 269 |
+
try:
|
| 270 |
+
score_mean, recall_mean, recall_score_mean, nonrecall_pred, precision = validScoreF1(score)
|
| 271 |
+
except:
|
| 272 |
+
breakpoint()
|
| 273 |
+
task2score[ll['Task Type'].strip()] += score_mean
|
| 274 |
+
task2recall[ll['Task Type'].strip()] += recall_mean
|
| 275 |
+
task2recall_score[ll['Task Type'].strip()] += recall_score_mean
|
| 276 |
+
task2nonrecall_pred[ll['Task Type'].strip()] += nonrecall_pred
|
| 277 |
+
task2precision[ll['Task Type'].strip()] += precision
|
| 278 |
+
|
| 279 |
+
|
| 280 |
+
|
| 281 |
+
for k,v in task2score.items():
|
| 282 |
+
if task2number[k] == 0:
|
| 283 |
+
task2score[k] = 0
|
| 284 |
+
task2recall[k] = 0
|
| 285 |
+
task2recall_score[k] = 0
|
| 286 |
+
task2nonrecall_pred[k] = 0
|
| 287 |
+
task2precision[k] = 0
|
| 288 |
+
else:
|
| 289 |
+
task2score[k] = v / task2number[k] * 100
|
| 290 |
+
task2recall[k] = task2recall[k] / task2number[k] * 100
|
| 291 |
+
task2recall_score[k] = task2recall_score[k] / task2number[k] * 100
|
| 292 |
+
# task2nonrecall_pred[k] = task2nonrecall_pred[k] / task2number[k]
|
| 293 |
+
task2precision[k] = task2precision[k] / task2number[k] * 100
|
| 294 |
+
|
| 295 |
+
print(json.dumps(task2number, indent=4))
|
| 296 |
+
print(json.dumps({k: round(v, 2) for k,v in task2score.items()}, indent=4))
|
| 297 |
+
print(json.dumps(task2recall, indent=4))
|
| 298 |
+
print(json.dumps(task2recall_score, indent=4))
|
| 299 |
+
|
| 300 |
+
print("Total question number: ", sum(task2number.values()))
|
| 301 |
+
|
| 302 |
+
print(f"Average Score: {sum(task2score.values())/len(task2number.values())}")
|
| 303 |
+
|
| 304 |
+
|
| 305 |
+
print(f"Average Recall: {sum(task2recall.values())/len(task2number.values())}")
|
| 306 |
+
print(f"Average Recall Score: {sum(task2recall_score.values())/len(task2number.values())}")
|
| 307 |
+
print(f"Average Precision: {sum(task2precision.values())/len(task2number.values())}")
|
| 308 |
+
|
| 309 |
+
|
| 310 |
+
|
| 311 |
+
print(f"Total Question Number: {sum(task2number.values())}")
|
| 312 |
+
print(f"Average FPS: {total_fps/total_response_number}")
|
| 313 |
+
print(f"Average KV Cache: {total_kv_cache_size/total_response_number}")
|
| 314 |
+
print(f"Average Non-Recall Pred: {sum(task2nonrecall_pred.values())/sum(task2number.values())}")
|
| 315 |
+
print(f"Average Response Number: {total_response_number/sum(task2number.values())}")
|
| 316 |
+
print(f"Average Answer Number: {total_answer_number/sum(task2number.values())}")
|
| 317 |
+
|
| 318 |
+
|
| 319 |
+
def generate_latex_table_row():
|
| 320 |
+
# Generate LaTeX format table row
|
| 321 |
+
print("\n# LaTeX format table row")
|
| 322 |
+
print(f"& {eval_model} ", end="")
|
| 323 |
+
|
| 324 |
+
# First 8 tasks
|
| 325 |
+
for task in list(task2number.keys())[:8]:
|
| 326 |
+
print(f"& {task2score[task]:.1f} ", end="")
|
| 327 |
+
|
| 328 |
+
|
| 329 |
+
# Overall average
|
| 330 |
+
print(f"& {sum(task2score.values())/len(task2number.values()):.1f}")
|
| 331 |
+
|
| 332 |
+
generate_latex_table_row()
|
| 333 |
+
|
| 334 |
+
def generate_latex_table_row():
|
| 335 |
+
# Generate LaTeX format table row
|
| 336 |
+
print("\n# LaTeX format table row")
|
| 337 |
+
print(f"& {eval_model} ", end="")
|
| 338 |
+
|
| 339 |
+
# First 8 tasks
|
| 340 |
+
for task in list(task2number.keys())[:8]:
|
| 341 |
+
print(f"& {task2recall[task]:.1f} ", end="")
|
| 342 |
+
|
| 343 |
+
|
| 344 |
+
# Overall average
|
| 345 |
+
print(f"& {sum(task2recall.values())/len(task2number.values()):.1f}")
|
| 346 |
+
|
| 347 |
+
generate_latex_table_row()
|
ESTP-Bench/estp_dataset/benchmark/eval_findcase.py
ADDED
|
@@ -0,0 +1,446 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import json,os
|
| 2 |
+
import numpy as np
|
| 3 |
+
|
| 4 |
+
def load_multiple_json(file_path):
|
| 5 |
+
"""读取包含多个 JSON 对象的文件，并将每个 JSON 对象解析成 Python 对象，存放在列表中。"""
|
| 6 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 7 |
+
content = f.read()
|
| 8 |
+
|
| 9 |
+
decoder = json.JSONDecoder()
|
| 10 |
+
pos = 0
|
| 11 |
+
results = []
|
| 12 |
+
content_length = len(content)
|
| 13 |
+
|
| 14 |
+
while pos < content_length:
|
| 15 |
+
# 跳过空白字符
|
| 16 |
+
while pos < content_length and content[pos].isspace():
|
| 17 |
+
pos += 1
|
| 18 |
+
if pos >= content_length:
|
| 19 |
+
break
|
| 20 |
+
try:
|
| 21 |
+
obj, new_pos = decoder.raw_decode(content, pos)
|
| 22 |
+
results.append(obj)
|
| 23 |
+
pos = new_pos
|
| 24 |
+
except json.JSONDecodeError as e:
|
| 25 |
+
# 出现解析错误则退出循环
|
| 26 |
+
print(f"JSON 解析错误: {e}")
|
| 27 |
+
break
|
| 28 |
+
return results
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
import argparse
|
| 32 |
+
|
| 33 |
+
def parse_args():
|
| 34 |
+
parser = argparse.ArgumentParser()
|
| 35 |
+
parser.add_argument('--eval_file', type=str, default='')
|
| 36 |
+
parser.add_argument('--eval_mode', type=str, default='all')
|
| 37 |
+
return parser.parse_args()
|
| 38 |
+
|
| 39 |
+
args = parse_args()
|
| 40 |
+
|
| 41 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_llama_5_5.json' # tmp_predict_VideollmOnline_v2_correctness
|
| 42 |
+
eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passiveevaluator_deepseek_1_2.json'
|
| 43 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5casesevaluator_deepseek_5_5.json'
|
| 44 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_5_5.json'
|
| 45 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/MiniCPMV_fbf_5cases_0.175evaluator_deepseek_1_2.json'
|
| 46 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 47 |
+
# eval_model = 'MiniCPMV'
|
| 48 |
+
|
| 49 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.9evaluator_deepseek_1_2.json'
|
| 50 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.8evaluator_deepseek_1_2.json'
|
| 51 |
+
# # eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LIVE_IT0.95.json'
|
| 52 |
+
# eval_model = 'VideollmOnline' # VideollmOnline MiniCPMV
|
| 53 |
+
|
| 54 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/MMDuetevaluator_deepseek_1_2.json'
|
| 55 |
+
# eval_model = 'MMDuet'
|
| 56 |
+
|
| 57 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_v6_fusion_dinov2evaluator_llama_5_5.json'
|
| 58 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estp_bench_sq_EWO_frame_by_frame_fitVal_5_cases_v2evaluator_llama_5_5.json'
|
| 59 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3Firstevaluator_llama_5_5.json'
|
| 60 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'
|
| 61 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq_fbf_EWOInDomainITstage2evaluator_llama_5_5.json'
|
| 62 |
+
|
| 63 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.6evaluator_llama_5_5.json'
|
| 64 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3First0.7evaluator_llama_5_5.json'
|
| 65 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3highv2evaluator_llama_5_5.json'
|
| 66 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3firstv2evaluator_llama_5_5.json'
|
| 67 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/result/estpBenchSq5Cases_fbf_EWOFitValStage3HighRegion_evaluator_llama_5_5.json'
|
| 68 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/train_/estpBenchSq5Cases_fbf_beaconlivel_h_stage2_v2evaluator_deepseek_5_5.json'
|
| 69 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 70 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage2evaluator_deepseek_5_5.json'
|
| 71 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high0.31_11evaluator_deepseek_1_2.json'
|
| 72 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 73 |
+
|
| 74 |
+
eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high11_evaluator_deepseek_1_2.json'
|
| 75 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage2evaluator_deepseek_1_2.json'
|
| 76 |
+
eval_model = 'EWO'
|
| 77 |
+
|
| 78 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_passiveevaluator_deepseek_1_2.json'
|
| 79 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 80 |
+
|
| 81 |
+
# eval_model = 'LLaVAOneVision'
|
| 82 |
+
|
| 83 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passiveevaluator_deepseek_1_2.json'
|
| 84 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 85 |
+
# eval_model = 'LLaVANextVideo7B'
|
| 86 |
+
|
| 87 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline_5cases/Qwen2VL_fbf_5casesevaluator_deepseek_5_5.json'
|
| 88 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175evaluator_deepseek_1_2.json'
|
| 89 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_passiveevaluator_deepseek_1_2.json'
|
| 90 |
+
# eval_model = 'Qwen2VL'
|
| 91 |
+
|
| 92 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json'
|
| 93 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_passiveevaluator_deepseek_1_2.json'
|
| 94 |
+
# eval_model = 'InternVLV28'
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
# eval_model = 'Lavila'
|
| 98 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/Lavila_streaming_v2evaluator_deepseek_1_2.json'
|
| 99 |
+
|
| 100 |
+
# eval_model = 'EgoVLP'
|
| 101 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json'
|
| 102 |
+
|
| 103 |
+
# eval_model = 'CLIP'
|
| 104 |
+
# eval_file = '/root/videollm-online/data/estp_dataset/estpSqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json'
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
parent_dir = os.path.dirname(eval_file)
|
| 108 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(eval_file.split('/')[-1])]
|
| 109 |
+
eval_result = []
|
| 110 |
+
for eval_file in eval_files:
|
| 111 |
+
eval_result += load_multiple_json(eval_file)
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
task2number = {
|
| 115 |
+
"Object Recognition": 0,
|
| 116 |
+
"Attribute Perception": 0,
|
| 117 |
+
"Text-Rich Understanding": 0,
|
| 118 |
+
"Object Localization": 0,
|
| 119 |
+
"Object State Change Recognition": 0,
|
| 120 |
+
"Ego Object Localization": 0,
|
| 121 |
+
"Ego Object State Change Recognition": 0,
|
| 122 |
+
"Action Recognition": 0,
|
| 123 |
+
"Object Function": 0,
|
| 124 |
+
"Information Function": 0,
|
| 125 |
+
"Action Reasoning": 0,
|
| 126 |
+
"Task Understanding": 0,
|
| 127 |
+
|
| 128 |
+
}
|
| 129 |
+
|
| 130 |
+
task2score = {
|
| 131 |
+
"Object Recognition": 0,
|
| 132 |
+
"Attribute Perception": 0,
|
| 133 |
+
"Text-Rich Understanding": 0,
|
| 134 |
+
"Object Localization": 0,
|
| 135 |
+
"Object State Change Recognition": 0,
|
| 136 |
+
"Ego Object Localization": 0,
|
| 137 |
+
"Ego Object State Change Recognition": 0,
|
| 138 |
+
"Action Recognition": 0,
|
| 139 |
+
"Object Function": 0,
|
| 140 |
+
"Information Function": 0,
|
| 141 |
+
"Action Reasoning": 0,
|
| 142 |
+
"Task Understanding": 0,
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
}
|
| 146 |
+
|
| 147 |
+
task2recall = {
|
| 148 |
+
"Object Recognition": 0,
|
| 149 |
+
"Attribute Perception": 0,
|
| 150 |
+
"Text-Rich Understanding": 0,
|
| 151 |
+
"Object Localization": 0,
|
| 152 |
+
"Object State Change Recognition": 0,
|
| 153 |
+
"Ego Object Localization": 0,
|
| 154 |
+
"Ego Object State Change Recognition": 0,
|
| 155 |
+
"Action Recognition": 0,
|
| 156 |
+
"Object Function": 0,
|
| 157 |
+
"Information Function": 0,
|
| 158 |
+
"Action Reasoning": 0,
|
| 159 |
+
"Task Understanding": 0,
|
| 160 |
+
|
| 161 |
+
|
| 162 |
+
}
|
| 163 |
+
|
| 164 |
+
task2recall_score = {
|
| 165 |
+
"Object Recognition": 0,
|
| 166 |
+
"Attribute Perception": 0,
|
| 167 |
+
"Text-Rich Understanding": 0,
|
| 168 |
+
"Object Localization": 0,
|
| 169 |
+
"Object State Change Recognition": 0,
|
| 170 |
+
"Ego Object Localization": 0,
|
| 171 |
+
"Ego Object State Change Recognition": 0,
|
| 172 |
+
"Action Recognition": 0,
|
| 173 |
+
"Object Function": 0,
|
| 174 |
+
"Information Function": 0,
|
| 175 |
+
"Action Reasoning": 0,
|
| 176 |
+
"Task Understanding": 0,
|
| 177 |
+
|
| 178 |
+
|
| 179 |
+
}
|
| 180 |
+
|
| 181 |
+
task2nonrecall_pred = {
|
| 182 |
+
"Object Recognition": 0,
|
| 183 |
+
"Attribute Perception": 0,
|
| 184 |
+
"Text-Rich Understanding": 0,
|
| 185 |
+
"Object Localization": 0,
|
| 186 |
+
"Object State Change Recognition": 0,
|
| 187 |
+
"Ego Object Localization": 0,
|
| 188 |
+
"Ego Object State Change Recognition": 0,
|
| 189 |
+
"Action Recognition": 0,
|
| 190 |
+
"Object Function": 0,
|
| 191 |
+
"Information Function": 0,
|
| 192 |
+
"Action Reasoning": 0,
|
| 193 |
+
"Task Understanding": 0,
|
| 194 |
+
|
| 195 |
+
}
|
| 196 |
+
|
| 197 |
+
task2precision = {
|
| 198 |
+
"Object Recognition": 0,
|
| 199 |
+
"Attribute Perception": 0,
|
| 200 |
+
"Text-Rich Understanding": 0,
|
| 201 |
+
"Object Localization": 0,
|
| 202 |
+
"Object State Change Recognition": 0,
|
| 203 |
+
"Ego Object Localization": 0,
|
| 204 |
+
"Ego Object State Change Recognition": 0,
|
| 205 |
+
"Action Recognition": 0,
|
| 206 |
+
"Object Function": 0,
|
| 207 |
+
"Information Function": 0,
|
| 208 |
+
"Action Reasoning": 0,
|
| 209 |
+
"Task Understanding": 0,
|
| 210 |
+
}
|
| 211 |
+
|
| 212 |
+
def scoreMean(score):
|
| 213 |
+
score = score.max(axis=0)
|
| 214 |
+
return score.mean()
|
| 215 |
+
|
| 216 |
+
|
| 217 |
+
def validScoreMean(score):
|
| 218 |
+
# FP
|
| 219 |
+
nonrecall_pred = (score.max(axis=0) == 0).sum()
|
| 220 |
+
|
| 221 |
+
# valid_score
|
| 222 |
+
valid_score = np.zeros(score.shape[0])
|
| 223 |
+
valid_recall = np.zeros(score.shape[0])
|
| 224 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 225 |
+
for i,s in enumerate(score):
|
| 226 |
+
valid_s = s[s > 0]
|
| 227 |
+
if len(valid_s) > 0:
|
| 228 |
+
valid_score[i] = valid_s.sum() / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 229 |
+
valid_score[i] = valid_s.sum() / (valid_s.sum() + (nonrecall_pred / score.shape[0]))
|
| 230 |
+
# valid_score[i] = valid_s.max() / (1 + (nonrecall_pred / score.shape[0]))
|
| 231 |
+
# valid_score[i] = valid_s.max() * len(valid_s) / (len(valid_s) + (nonrecall_pred / score.shape[0]))
|
| 232 |
+
|
| 233 |
+
valid_recall[i] = 1
|
| 234 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 235 |
+
return valid_score.mean(), valid_recall.mean(), valid_recall_score.mean(), nonrecall_pred
|
| 236 |
+
|
| 237 |
+
BETA = 1
|
| 238 |
+
def validScoreF1(score):
|
| 239 |
+
|
| 240 |
+
if score.shape[1] == 0:
|
| 241 |
+
return 0, 0, 0, 0, 0
|
| 242 |
+
# FP
|
| 243 |
+
FP = (score.max(axis=0) == 0).sum()
|
| 244 |
+
# TP
|
| 245 |
+
TP = 0
|
| 246 |
+
|
| 247 |
+
# valid_score
|
| 248 |
+
valid_score = np.zeros(score.shape[0])
|
| 249 |
+
valid_recall = np.zeros(score.shape[0])
|
| 250 |
+
valid_recall_score = np.zeros(score.shape[0])
|
| 251 |
+
for i,s in enumerate(score):
|
| 252 |
+
valid_s = s[s > 0]
|
| 253 |
+
if len(valid_s) > 0:
|
| 254 |
+
# four type compute text-time precision
|
| 255 |
+
valid_score[i] = valid_s.sum() / len(valid_s)
|
| 256 |
+
# valid_score[i] = valid_s.max()
|
| 257 |
+
# valid_score[i] = valid_s.sum()
|
| 258 |
+
|
| 259 |
+
valid_recall[i] = 1
|
| 260 |
+
valid_recall_score[i] = valid_s.sum() / len(valid_s)
|
| 261 |
+
# valid_recall_score[i] = valid_s.max()
|
| 262 |
+
# valid_score[i] = valid_s.sum()
|
| 263 |
+
|
| 264 |
+
TP = valid_score.sum()
|
| 265 |
+
precision = TP / (TP + FP)
|
| 266 |
+
recall = valid_recall_score.mean()
|
| 267 |
+
if precision == 0 or recall == 0:
|
| 268 |
+
F1 = 0
|
| 269 |
+
else:
|
| 270 |
+
F1 = (1 + BETA**2) * precision * recall / ((BETA**2 * precision) + recall)
|
| 271 |
+
|
| 272 |
+
F1 = 2*TP / (2*TP + FP + score.shape[0] - valid_recall.sum())
|
| 273 |
+
return F1, valid_recall.mean(), valid_recall_score.mean(), FP, precision
|
| 274 |
+
|
| 275 |
+
def topkValidScoreMean(score, k=10):
|
| 276 |
+
score = score.max(axis=1)
|
| 277 |
+
score = score[:k]
|
| 278 |
+
valid_score = score[score > 0]
|
| 279 |
+
return valid_score.mean()
|
| 280 |
+
|
| 281 |
+
total_fps = 0
|
| 282 |
+
total_kv_cache_size = 0
|
| 283 |
+
total_response_number = 0
|
| 284 |
+
total_answer_number = 0
|
| 285 |
+
total_precision = 0
|
| 286 |
+
high_performance_cases = []
|
| 287 |
+
|
| 288 |
+
for ll in eval_result:
|
| 289 |
+
if eval_model in ll.keys():
|
| 290 |
+
total_answer_number += len(ll['conversation'])
|
| 291 |
+
this_turn_response_number = 0
|
| 292 |
+
for response in ll[eval_model]:
|
| 293 |
+
if response['role'].lower() == 'assistant':
|
| 294 |
+
if 'fps' in response:
|
| 295 |
+
total_fps+=response['fps']
|
| 296 |
+
total_response_number += 1
|
| 297 |
+
this_turn_response_number += 1
|
| 298 |
+
if 'kv_cache_size' in response:
|
| 299 |
+
total_kv_cache_size += response['kv_cache_size']
|
| 300 |
+
task2number[ll['Task Type'].strip()] += 1
|
| 301 |
+
# breakpoint()
|
| 302 |
+
text_score = np.array(ll['evaluator_output_text']) / 10
|
| 303 |
+
reponse_score = np.array(ll['evaluator_output_reponse']) / 10
|
| 304 |
+
# if ll['Task Type'] == 'Ego Object State Change Recognition':
|
| 305 |
+
if args.eval_mode == 'all' and eval_model not in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 306 |
+
score = (text_score+reponse_score)
|
| 307 |
+
elif args.eval_mode == 'text':
|
| 308 |
+
score = text_score
|
| 309 |
+
elif args.eval_mode == 'response' or eval_model in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 310 |
+
score = reponse_score
|
| 311 |
+
score_mean, recall_mean, recall_score_mean, nonrecall_pred, precision = validScoreF1(score)
|
| 312 |
+
|
| 313 |
+
# 保存性能较高且gt多于一个的case
|
| 314 |
+
if (score_mean > 0.5 and ll['Task Type'].strip() not in ['Task Understanding', 'Action Reasoning']) or score_mean > 0.7:
|
| 315 |
+
if score.shape[0] / score.shape[1] > 0.5 and score.shape[0] > 1:
|
| 316 |
+
high_performance_cases.append({
|
| 317 |
+
'case_id': ll.get('id', ''),
|
| 318 |
+
'task_type': ll['Task Type'].strip(),
|
| 319 |
+
'score_mean': float(score_mean),
|
| 320 |
+
'gt_count': int(np.sum(score > 0)),
|
| 321 |
+
'data': ll
|
| 322 |
+
})
|
| 323 |
+
|
| 324 |
+
task2score[ll['Task Type'].strip()] += score_mean
|
| 325 |
+
task2recall[ll['Task Type'].strip()] += recall_mean
|
| 326 |
+
task2recall_score[ll['Task Type'].strip()] += recall_score_mean
|
| 327 |
+
task2nonrecall_pred[ll['Task Type'].strip()] += nonrecall_pred
|
| 328 |
+
task2precision[ll['Task Type'].strip()] += precision
|
| 329 |
+
total_precision += nonrecall_pred / this_turn_response_number if this_turn_response_number > 0 else 0
|
| 330 |
+
|
| 331 |
+
# 将高性能案例保存到JSON文件
|
| 332 |
+
if high_performance_cases:
|
| 333 |
+
output_file = os.path.join(os.path.dirname(eval_file), f'high_performance_cases_{eval_model}.json')
|
| 334 |
+
with open(output_file, 'w', encoding='utf-8') as f:
|
| 335 |
+
json.dump(high_performance_cases, f, ensure_ascii=False, indent=2)
|
| 336 |
+
print(f"Saved {len(high_performance_cases)} high performance cases to {output_file}")
|
| 337 |
+
|
| 338 |
+
|
| 339 |
+
|
| 340 |
+
for k,v in task2score.items():
|
| 341 |
+
if task2number[k] == 0:
|
| 342 |
+
task2score[k] = 0
|
| 343 |
+
task2recall[k] = 0
|
| 344 |
+
task2recall_score[k] = 0
|
| 345 |
+
task2nonrecall_pred[k] = 0
|
| 346 |
+
task2precision[k] = 0
|
| 347 |
+
else:
|
| 348 |
+
task2score[k] = v / task2number[k] * 100
|
| 349 |
+
task2recall[k] = task2recall[k] / task2number[k] * 100
|
| 350 |
+
task2recall_score[k] = task2recall_score[k] / task2number[k] * 100
|
| 351 |
+
# task2nonrecall_pred[k] = task2nonrecall_pred[k] / task2number[k]
|
| 352 |
+
task2precision[k] = task2precision[k] / task2number[k] * 100
|
| 353 |
+
|
| 354 |
+
# print(json.dumps(task2number, indent=4))
|
| 355 |
+
# print(json.dumps({k: round(v, 2) for k,v in task2score.items()}, indent=4))
|
| 356 |
+
# print(json.dumps(task2recall, indent=4))
|
| 357 |
+
# print(json.dumps(task2recall_score, indent=4))
|
| 358 |
+
# print(json.dumps(task2nonrecall_pred, indent=4))
|
| 359 |
+
|
| 360 |
+
print("Total question number: ", sum(task2number.values()))
|
| 361 |
+
|
| 362 |
+
print(f"Average Score: {sum(task2score.values())/len(task2number.values())}")
|
| 363 |
+
|
| 364 |
+
# Calculate average for first 8 task types
|
| 365 |
+
first_8_tasks = list(task2number.keys())[:8]
|
| 366 |
+
first_8_scores = [task2score[task] for task in first_8_tasks]
|
| 367 |
+
first_8_valid_scores = [score for score in first_8_scores if score > 0]
|
| 368 |
+
first_8_recall_scores = [task2recall[task] for task in first_8_tasks]
|
| 369 |
+
try:
|
| 370 |
+
print(f"Average Score (First 8 Tasks): {sum(first_8_valid_scores)/len(first_8_valid_scores):.2f}")
|
| 371 |
+
except:
|
| 372 |
+
print(f"Average Score (First 8 Tasks): 0.0")
|
| 373 |
+
|
| 374 |
+
# Calculate average for last 4 task types
|
| 375 |
+
last_4_tasks = list(task2number.keys())[-4:]
|
| 376 |
+
last_4_scores = [task2score[task] for task in last_4_tasks]
|
| 377 |
+
last_4_valid_scores = [score for score in last_4_scores if score > 0]
|
| 378 |
+
last_4_recall_scores = [task2recall[task] for task in last_4_tasks]
|
| 379 |
+
try:
|
| 380 |
+
print(f"Average Score (Last 4 Tasks): {sum(last_4_valid_scores)/len(last_4_valid_scores):.2f}")
|
| 381 |
+
except:
|
| 382 |
+
print(f"Average Score (Last 4 Tasks): 0.0")
|
| 383 |
+
|
| 384 |
+
|
| 385 |
+
print(f"Average Recall: {sum(task2recall.values())/len(task2number.values())}")
|
| 386 |
+
print(f"Average Recall Score: {sum(task2recall_score.values())/len(task2number.values())}")
|
| 387 |
+
print(f"Average Precision: {sum(task2precision.values())/len(task2number.values())}")
|
| 388 |
+
|
| 389 |
+
|
| 390 |
+
print(f"Total Question Number: {sum(task2number.values())}")
|
| 391 |
+
print(f"Average FPS: {total_fps/total_response_number}")
|
| 392 |
+
print(f"Average KV Cache: {total_kv_cache_size/total_response_number}")
|
| 393 |
+
print(f"Average Non-Recall Pred: {sum(task2nonrecall_pred.values())/sum(task2number.values())}")
|
| 394 |
+
print(f"Average Response Number: {total_response_number/sum(task2number.values())}")
|
| 395 |
+
print(f"Average Precision: {total_precision/sum(task2number.values()) * 100}")
|
| 396 |
+
|
| 397 |
+
print(f"[{sum(task2recall.values())/len(task2number.values()):.1f}, {sum(task2precision.values())/len(task2number.values()):.1f}]")
|
| 398 |
+
|
| 399 |
+
def generate_latex_table_row():
|
| 400 |
+
# Generate LaTeX format table row
|
| 401 |
+
print("\n# LaTeX format table row")
|
| 402 |
+
print(f"& {eval_model} ", end="")
|
| 403 |
+
|
| 404 |
+
# First 8 tasks
|
| 405 |
+
for task in list(task2number.keys())[:8]:
|
| 406 |
+
print(f"& {task2score[task]:.1f} ", end="")
|
| 407 |
+
|
| 408 |
+
# Average of first 8 tasks
|
| 409 |
+
print(f"& {sum(first_8_valid_scores)/len(first_8_valid_scores):.1f} ", end="")
|
| 410 |
+
|
| 411 |
+
# Last 4 tasks
|
| 412 |
+
for task in list(task2number.keys())[-4:]:
|
| 413 |
+
print(f"& {task2score[task]:.1f} ", end="")
|
| 414 |
+
|
| 415 |
+
# Average of last 4 tasks
|
| 416 |
+
print(f"& {sum(last_4_valid_scores)/len(last_4_valid_scores):.1f} ", end="")
|
| 417 |
+
|
| 418 |
+
# Overall average
|
| 419 |
+
# print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")
|
| 420 |
+
|
| 421 |
+
generate_latex_table_row()
|
| 422 |
+
|
| 423 |
+
|
| 424 |
+
# def generate_latex_table_row():
|
| 425 |
+
# # Generate LaTeX format table row
|
| 426 |
+
# print("\n# LaTeX format table row")
|
| 427 |
+
# print(f"& {eval_model} ", end="")
|
| 428 |
+
|
| 429 |
+
# # First 8 tasks
|
| 430 |
+
# for task in list(task2number.keys())[:8]:
|
| 431 |
+
# print(f"& {task2recall[task]:.1f} ", end="")
|
| 432 |
+
|
| 433 |
+
# # Average of first 8 tasks
|
| 434 |
+
# print(f"& {sum(first_8_recall_scores)/len(first_8_recall_scores):.1f} ", end="")
|
| 435 |
+
|
| 436 |
+
# # Last 4 tasks
|
| 437 |
+
# for task in list(task2number.keys())[-4:]:
|
| 438 |
+
# print(f"& {task2recall[task]:.1f} ", end="")
|
| 439 |
+
|
| 440 |
+
# # Average of last 4 tasks
|
| 441 |
+
# print(f"& {sum(last_4_recall_scores)/len(last_4_recall_scores):.1f} ", end="")
|
| 442 |
+
|
| 443 |
+
# # Overall average
|
| 444 |
+
# # print(f"& {sum(task2score.values())/len(task2number.values()):.1f} \\\\")
|
| 445 |
+
|
| 446 |
+
# generate_latex_table_row()
|
ESTP-Bench/estp_dataset/benchmark/eval_singleQA.sh
ADDED
|
@@ -0,0 +1,387 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
######################################################## passive inference ########################################################
|
| 2 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 3 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 4 |
+
--pred_file /root/videollm-online/data/estp_dataset/result/estp_bench_sq_MiniCPMV_passive.json \
|
| 5 |
+
--eval_model MiniCPMV \
|
| 6 |
+
--concat True \
|
| 7 |
+
--anticipation 1 \
|
| 8 |
+
--latency 2 \
|
| 9 |
+
--evaluator_llm deepseek \
|
| 10 |
+
--master_port 2923 \
|
| 11 |
+
|
| 12 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 13 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 14 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_passive_v2.json \
|
| 15 |
+
--eval_model MiniCPMV \
|
| 16 |
+
--concat True \
|
| 17 |
+
--anticipation 1 \
|
| 18 |
+
--latency 2 \
|
| 19 |
+
--evaluator_llm deepseek \
|
| 20 |
+
--qa_type MultiQA \
|
| 21 |
+
--master_port 2923 \
|
| 22 |
+
|
| 23 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 24 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 25 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_passive.json \
|
| 26 |
+
--eval_model Qwen2VL \
|
| 27 |
+
--concat True \
|
| 28 |
+
--anticipation 1 \
|
| 29 |
+
--latency 2 \
|
| 30 |
+
--evaluator_llm deepseek \
|
| 31 |
+
--master_port 2918 \
|
| 32 |
+
|
| 33 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 34 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 35 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/Qwen2VL_passive_v2.json \
|
| 36 |
+
--eval_model Qwen2VL \
|
| 37 |
+
--concat True \
|
| 38 |
+
--anticipation 1 \
|
| 39 |
+
--latency 2 \
|
| 40 |
+
--evaluator_llm deepseek \
|
| 41 |
+
--qa_type MultiQA \
|
| 42 |
+
--master_port 2918 \
|
| 43 |
+
|
| 44 |
+
|
| 45 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 46 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 47 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_passive.json \
|
| 48 |
+
--eval_model LLaVAOneVision \
|
| 49 |
+
--anticipation 1 \
|
| 50 |
+
--latency 2 \
|
| 51 |
+
--evaluator_llm deepseek \
|
| 52 |
+
|
| 53 |
+
conda activate videollm
|
| 54 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 55 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 56 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVAOneVision_passive_v2.json \
|
| 57 |
+
--eval_model LLaVAOneVision \
|
| 58 |
+
--anticipation 1 \
|
| 59 |
+
--latency 2 \
|
| 60 |
+
--qa_type MultiQA \
|
| 61 |
+
--evaluator_llm deepseek \
|
| 62 |
+
--master_port 2298 \
|
| 63 |
+
|
| 64 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 65 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 66 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_passive.json \
|
| 67 |
+
--eval_model LLaVANextVideo7B \
|
| 68 |
+
--anticipation 1 \
|
| 69 |
+
--latency 2 \
|
| 70 |
+
--evaluator_llm deepseek \
|
| 71 |
+
--master_port 3980 \
|
| 72 |
+
|
| 73 |
+
conda activate videollm
|
| 74 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 75 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 76 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVANextVideo7B_passive_v2.json \
|
| 77 |
+
--eval_model LLaVANextVideo7B \
|
| 78 |
+
--anticipation 1 \
|
| 79 |
+
--latency 2 \
|
| 80 |
+
--qa_type MultiQA \
|
| 81 |
+
--evaluator_llm deepseek \
|
| 82 |
+
--master_port 3980 \
|
| 83 |
+
|
| 84 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 85 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 86 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_passive.json \
|
| 87 |
+
--eval_model InternVLV28 \
|
| 88 |
+
--anticipation 1 \
|
| 89 |
+
--latency 2 \
|
| 90 |
+
--evaluator_llm deepseek \
|
| 91 |
+
--master_port 3989 \
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 95 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 96 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/InternVLV28_passive_v2.json \
|
| 97 |
+
--eval_model InternVLV28 \
|
| 98 |
+
--anticipation 1 \
|
| 99 |
+
--latency 2 \
|
| 100 |
+
--qa_type MultiQA \
|
| 101 |
+
--evaluator_llm deepseek \
|
| 102 |
+
--master_port 3989 \
|
| 103 |
+
######################################################## turn ask on 5 cases ########################################################
|
| 104 |
+
|
| 105 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 106 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 107 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/MiniCPMV_fbf_singleQA_0.175_v2.json \
|
| 108 |
+
--eval_model MiniCPMV \
|
| 109 |
+
--concat True \
|
| 110 |
+
--anticipation 1 \
|
| 111 |
+
--latency 2 \
|
| 112 |
+
--evaluator_llm deepseek \
|
| 113 |
+
--master_port 2951 \
|
| 114 |
+
|
| 115 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 116 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 117 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/MiniCPMV_fbf_0.175.json \
|
| 118 |
+
--eval_model MiniCPMV \
|
| 119 |
+
--concat True \
|
| 120 |
+
--anticipation 1 \
|
| 121 |
+
--latency 2 \
|
| 122 |
+
--qa_type MultiQA \
|
| 123 |
+
--evaluator_llm deepseek \
|
| 124 |
+
--master_port 3951 \
|
| 125 |
+
|
| 126 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 127 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 128 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/Qwen2VL_fbf_singleQA_0.175.json \
|
| 129 |
+
--eval_model Qwen2VL \
|
| 130 |
+
--concat True \
|
| 131 |
+
--anticipation 1 \
|
| 132 |
+
--latency 2 \
|
| 133 |
+
--evaluator_llm deepseek \
|
| 134 |
+
--master_port 2951 \
|
| 135 |
+
|
| 136 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 137 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 138 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/Qwen2VL_fbf_0.175.json \
|
| 139 |
+
--eval_model Qwen2VL \
|
| 140 |
+
--concat True \
|
| 141 |
+
--anticipation 1 \
|
| 142 |
+
--latency 2 \
|
| 143 |
+
--qa_type MultiQA \
|
| 144 |
+
--evaluator_llm deepseek \
|
| 145 |
+
--master_port 2151 \
|
| 146 |
+
|
| 147 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 148 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 149 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVAOneVision_fbf_singleQA_0.175.json \
|
| 150 |
+
--eval_model LLaVAOneVision \
|
| 151 |
+
--concat True \
|
| 152 |
+
--anticipation 1 \
|
| 153 |
+
--latency 2 \
|
| 154 |
+
--evaluator_llm deepseek \
|
| 155 |
+
--master_port 2951 \
|
| 156 |
+
|
| 157 |
+
conda activate videollm
|
| 158 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 159 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 160 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVAOneVision_fbf_0.175.json \
|
| 161 |
+
--eval_model LLaVAOneVision \
|
| 162 |
+
--concat True \
|
| 163 |
+
--anticipation 1 \
|
| 164 |
+
--latency 2 \
|
| 165 |
+
--qa_type MultiQA \
|
| 166 |
+
--evaluator_llm deepseek \
|
| 167 |
+
--master_port 2951 \
|
| 168 |
+
|
| 169 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 170 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 171 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/InternVLV28_fbf_0.175.json \
|
| 172 |
+
--eval_model InternVLV28 \
|
| 173 |
+
--concat True \
|
| 174 |
+
--anticipation 1 \
|
| 175 |
+
--latency 2 \
|
| 176 |
+
--evaluator_llm deepseek \
|
| 177 |
+
--master_port 1951 \
|
| 178 |
+
|
| 179 |
+
|
| 180 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 181 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 182 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json \
|
| 183 |
+
--eval_model InternVLV28 \
|
| 184 |
+
--concat True \
|
| 185 |
+
--anticipation 1 \
|
| 186 |
+
--latency 2 \
|
| 187 |
+
--qa_type MultiQA \
|
| 188 |
+
--evaluator_llm deepseek \
|
| 189 |
+
--master_port 2951 \
|
| 190 |
+
|
| 191 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 192 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 193 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/LLaVANextVideo7B_fbf_singleQA_0.175.json \
|
| 194 |
+
--eval_model LLaVANextVideo7B \
|
| 195 |
+
--concat True \
|
| 196 |
+
--anticipation 1 \
|
| 197 |
+
--latency 2 \
|
| 198 |
+
--evaluator_llm deepseek \
|
| 199 |
+
--master_port 1921 \
|
| 200 |
+
|
| 201 |
+
conda activate videollm
|
| 202 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 203 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 204 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/LLaVANextVideo7B_fbf_0.175.json \
|
| 205 |
+
--eval_model LLaVANextVideo7B \
|
| 206 |
+
--concat True \
|
| 207 |
+
--anticipation 1 \
|
| 208 |
+
--latency 2 \
|
| 209 |
+
--qa_type MultiQA \
|
| 210 |
+
--evaluator_llm deepseek \
|
| 211 |
+
--master_port 3951 \
|
| 212 |
+
|
| 213 |
+
######################################################## streaming inference ########################################################
|
| 214 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 215 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 216 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/EgoVLP_streaming_v2.json \
|
| 217 |
+
--eval_model EgoVLP \
|
| 218 |
+
--anticipation 1 \
|
| 219 |
+
--latency 2 \
|
| 220 |
+
--evaluator_llm deepseek \
|
| 221 |
+
--master_port 2851 \
|
| 222 |
+
|
| 223 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 224 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 225 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2.json \
|
| 226 |
+
--eval_model EgoVLP \
|
| 227 |
+
--anticipation 1 \
|
| 228 |
+
--latency 2 \
|
| 229 |
+
--qa_type MultiQA \
|
| 230 |
+
--evaluator_llm deepseek \
|
| 231 |
+
--master_port 2951 \
|
| 232 |
+
|
| 233 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 234 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 235 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/Lavila_streaming_v2.json \
|
| 236 |
+
--eval_model Lavila \
|
| 237 |
+
--anticipation 1 \
|
| 238 |
+
--latency 2 \
|
| 239 |
+
--evaluator_llm deepseek \
|
| 240 |
+
|
| 241 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 242 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 243 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/Lavila_streaming_v2.json \
|
| 244 |
+
--eval_model Lavila \
|
| 245 |
+
--anticipation 1 \
|
| 246 |
+
--latency 2 \
|
| 247 |
+
--qa_type MultiQA \
|
| 248 |
+
--evaluator_llm deepseek \
|
| 249 |
+
--master_port 2951 \
|
| 250 |
+
|
| 251 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 252 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 253 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/CLIP_streaming_v2.json \
|
| 254 |
+
--eval_model CLIP \
|
| 255 |
+
--anticipation 1 \
|
| 256 |
+
--latency 2 \
|
| 257 |
+
--evaluator_llm deepseek \
|
| 258 |
+
--master_port 2551 \
|
| 259 |
+
|
| 260 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 261 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 262 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/CLIP_streaming_v2.json \
|
| 263 |
+
--eval_model CLIP \
|
| 264 |
+
--anticipation 1 \
|
| 265 |
+
--latency 2 \
|
| 266 |
+
--evaluator_llm deepseek \
|
| 267 |
+
--qa_type MultiQA \
|
| 268 |
+
--master_port 2651 \
|
| 269 |
+
|
| 270 |
+
######################################################## online inference ########################################################
|
| 271 |
+
|
| 272 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 273 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 274 |
+
--pred_file /root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.8.json \
|
| 275 |
+
--eval_model VideollmOnline \
|
| 276 |
+
--concat True \
|
| 277 |
+
--anticipation 1 \
|
| 278 |
+
--latency 2 \
|
| 279 |
+
--evaluator_llm deepseek \
|
| 280 |
+
|
| 281 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 282 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 283 |
+
--pred_file /root/videollm-online/data/estp_dataset/estp_bench_sq_VideollmOnline0.9.json \
|
| 284 |
+
--eval_model VideollmOnline \
|
| 285 |
+
--concat True \
|
| 286 |
+
--anticipation 1 \
|
| 287 |
+
--latency 2 \
|
| 288 |
+
--evaluator_llm deepseek \
|
| 289 |
+
--master_port 3019
|
| 290 |
+
|
| 291 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 292 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 293 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/VideollmOnline0.8.json \
|
| 294 |
+
--eval_model VideollmOnline \
|
| 295 |
+
--concat True \
|
| 296 |
+
--anticipation 1 \
|
| 297 |
+
--latency 2 \
|
| 298 |
+
--qa_type MultiQA \
|
| 299 |
+
--evaluator_llm deepseek \
|
| 300 |
+
--master_port 3019
|
| 301 |
+
|
| 302 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 303 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 304 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_baseline/MMDuet.json \
|
| 305 |
+
--eval_model MMDuet \
|
| 306 |
+
--concat True \
|
| 307 |
+
--anticipation 1 \
|
| 308 |
+
--latency 2 \
|
| 309 |
+
--evaluator_llm deepseek \
|
| 310 |
+
--master_port 3019
|
| 311 |
+
|
| 312 |
+
conda activate videollm
|
| 313 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 314 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 315 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_baseline/MMDuet.json \
|
| 316 |
+
--eval_model MMDuet \
|
| 317 |
+
--concat True \
|
| 318 |
+
--anticipation 1 \
|
| 319 |
+
--latency 2 \
|
| 320 |
+
--qa_type MultiQA \
|
| 321 |
+
--evaluator_llm deepseek \
|
| 322 |
+
--master_port 3019
|
| 323 |
+
|
| 324 |
+
######################################################## ours inference ########################################################
|
| 325 |
+
|
| 326 |
+
|
| 327 |
+
# single QA
|
| 328 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 329 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 330 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage2.json \
|
| 331 |
+
--eval_model EWO \
|
| 332 |
+
--concat True \
|
| 333 |
+
--anticipation 5 \
|
| 334 |
+
--latency 5 \
|
| 335 |
+
--evaluator_llm deepseek \
|
| 336 |
+
--master_port 2900 \
|
| 337 |
+
|
| 338 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 339 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 340 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high11_.json \
|
| 341 |
+
--eval_model EWO \
|
| 342 |
+
--concat True \
|
| 343 |
+
--anticipation 1 \
|
| 344 |
+
--latency 5 \
|
| 345 |
+
--evaluator_llm deepseek \
|
| 346 |
+
--master_port 2801 \
|
| 347 |
+
|
| 348 |
+
|
| 349 |
+
# subset 5 cases
|
| 350 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 351 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 352 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage2.json \
|
| 353 |
+
--eval_model EWO \
|
| 354 |
+
--concat True \
|
| 355 |
+
--anticipation 5 \
|
| 356 |
+
--latency 5 \
|
| 357 |
+
--evaluator_llm deepseek \
|
| 358 |
+
|
| 359 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 360 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 361 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_ours/LivebaseStage3_high0.31_11_low.json \
|
| 362 |
+
--eval_model EWO \
|
| 363 |
+
--concat True \
|
| 364 |
+
--anticipation 1 \
|
| 365 |
+
--latency 2 \
|
| 366 |
+
--evaluator_llm deepseek \
|
| 367 |
+
|
| 368 |
+
|
| 369 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 370 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 371 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpSqa_ours_5cases/LivebaseStage3_high0.31_1_low.json \
|
| 372 |
+
--eval_model EWO \
|
| 373 |
+
--concat True \
|
| 374 |
+
--anticipation 1 \
|
| 375 |
+
--latency 2 \
|
| 376 |
+
--evaluator_llm deepseek \
|
| 377 |
+
|
| 378 |
+
# multi QA
|
| 379 |
+
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
|
| 380 |
+
python /root/videollm-online/data/estp_dataset/benchmark/evalate_singleQA.py \
|
| 381 |
+
--pred_file /root/videollm-online/data/estp_dataset/estpCqa_ours/LivebaseStage2_low.json \
|
| 382 |
+
--eval_model EWO \
|
| 383 |
+
--concat True \
|
| 384 |
+
--anticipation 1 \
|
| 385 |
+
--qa_type MultiQA \
|
| 386 |
+
--latency 2 \
|
| 387 |
+
--evaluator_llm deepseek \
|
ESTP-Bench/estp_dataset/benchmark/evalate_singleQA.py
ADDED
|
@@ -0,0 +1,484 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os, sys, re, requests, random
|
| 2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
+
import json
|
| 4 |
+
from tqdm import tqdm
|
| 5 |
+
import argparse
|
| 6 |
+
import torch
|
| 7 |
+
import numpy as np
|
| 8 |
+
import pandas as pd
|
| 9 |
+
import torch.distributed as dist
|
| 10 |
+
import torch.multiprocessing as mp
|
| 11 |
+
import openai
|
| 12 |
+
|
| 13 |
+
import math
|
| 14 |
+
def ceil_time_by_fps(time: float, fps: int, min_time: float, max_time: float):
|
| 15 |
+
return min(max(math.ceil(time * fps) / fps, min_time), max_time)
|
| 16 |
+
|
| 17 |
+
EVALUATOR_PROMPT = [
|
| 18 |
+
{"role": "system", "content": (
|
| 19 |
+
"You are an evaluator for a video question answering system. Your task is to rate the "
|
| 20 |
+
"correctness of the predicted answers against the ground truth answers. Use the following scale to assign a score:\n"
|
| 21 |
+
"- 5: Perfect match; the predicted answer is completely correct and contains all the relevant information.\n"
|
| 22 |
+
"- 4: Mostly correct; the predicted answer is largely accurate but may have minor omissions or slight inaccuracies.\n"
|
| 23 |
+
"- 3: Partially correct; the predicted answer has some correct information, but also contains significant inaccuracies or missing key points.\n"
|
| 24 |
+
"- 2: Slightly correct; the predicted answer has only a few correct elements, but most of the information is incorrect or irrelevant, or the predicted answer conflicts with the ground truth answer.\n"
|
| 25 |
+
"- 1: Incorrect; the predicted answer is entirely wrong or does not address the question at all.\n\n"
|
| 26 |
+
"Here are some examples to guide you:")
|
| 27 |
+
},
|
| 28 |
+
{"role": "user", "content": "Question: How can I achieve my goal, such as stir-frying the ingredients, step by step? Can you explain each stage?\nGround Truth Answer: Let\u2019s begin! First, add oil to the pan to prepare for cooking.\nPredicted Answer: Heat oil in the pan."},
|
| 29 |
+
{"role": "assistant", "content": "2"},
|
| 30 |
+
|
| 31 |
+
{"role": "user", "content": "Question: what is the category of the object I hold?\nGround Truth Answer: The object you hold is a computer mouse.\nPredicted Answer: It is a computer mouse."},
|
| 32 |
+
{"role": "assistant", "content": "5"},
|
| 33 |
+
|
| 34 |
+
{"role": "user", "content": "Question: Can you remind me how the towel changes position when I dip it in the water?\nGround Truth Answer: The towel is initially held in your right hand, and you dip it into a purple plastic basin filled with water. The towel becomes submerged, absorbing water, and is then lifted out, allowing excess water to drip off.\nPredicted Answer: The towel changes position when you dip it in the water."},
|
| 35 |
+
{"role": "assistant", "content": "3"},
|
| 36 |
+
|
| 37 |
+
{"role": "user", "content": "Question: Can you remind me when the state of the trunk changes? \nGround Truth Answer: The trunk is opened by the observer.\nPredicted Answer: The trunk is closed."},
|
| 38 |
+
{"role": "assistant", "content": "1"},
|
| 39 |
+
|
| 40 |
+
{"role": "user", "content": "Question: Where're the cars location?\nGround Truth Answer: The cars are one the road far beneath you.\nPredicted Answer: The cars are located at the base of the cliff."},
|
| 41 |
+
{"role": "assistant", "content": "3"},
|
| 42 |
+
]
|
| 43 |
+
|
| 44 |
+
class CorrectnessEvaluator:
|
| 45 |
+
@torch.no_grad()
|
| 46 |
+
def __init__(self, llm_pretrained):
|
| 47 |
+
self.tokenizer = AutoTokenizer.from_pretrained(llm_pretrained)
|
| 48 |
+
self.model = AutoModelForCausalLM.from_pretrained(llm_pretrained, torch_dtype=torch.bfloat16, device_map='auto')
|
| 49 |
+
conversation = EVALUATOR_PROMPT
|
| 50 |
+
|
| 51 |
+
prompt_input = self.tokenizer.apply_chat_template(conversation, return_tensors='pt', return_dict=True).to(self.model.device)
|
| 52 |
+
outputs = self.model(**prompt_input, use_cache=True)
|
| 53 |
+
self.prompt_past_key_values = outputs.past_key_values
|
| 54 |
+
self.prompt_input_ids = prompt_input.input_ids
|
| 55 |
+
|
| 56 |
+
@torch.no_grad()
|
| 57 |
+
def evaluate(self, question, gold_answer, pred_answer):
|
| 58 |
+
conversation = [
|
| 59 |
+
{"role": "user", "content": f"Question: {question}\nGround Truth Answer: {gold_answer}\nPredicted Answer: {pred_answer}"},
|
| 60 |
+
{"role": "assistant", "content": ""}
|
| 61 |
+
]
|
| 62 |
+
new_input_ids = self.tokenizer.apply_chat_template(conversation, return_tensors='pt').to(self.model.device)
|
| 63 |
+
first_eot_index = (new_input_ids == 128009).nonzero()[0, -1] # remove the system prompt before the user turn (i.e., the first turn) of llama tokenizer
|
| 64 |
+
new_input_ids = new_input_ids[:, first_eot_index + 1:-1] # -1 (the last token): '<|eot|>'
|
| 65 |
+
|
| 66 |
+
all_input_ids = torch.cat([self.prompt_input_ids, new_input_ids], dim=1)
|
| 67 |
+
generated_ids = self.model.generate(input_ids=all_input_ids, use_cache=True, max_new_tokens=32)
|
| 68 |
+
generated_ids = generated_ids[:, all_input_ids.size(1):]
|
| 69 |
+
decoded_text = self.tokenizer.decode(generated_ids[0], skip_special_tokens=True)
|
| 70 |
+
score = int(decoded_text[0]) if decoded_text[0] in '12345' else 1
|
| 71 |
+
return score
|
| 72 |
+
|
| 73 |
+
class GPTCorrectnessEvaluator:
|
| 74 |
+
def __init__(self, api_key):
|
| 75 |
+
self.client = openai.OpenAI(api_key=api_key)
|
| 76 |
+
self.conversation = EVALUATOR_PROMPT
|
| 77 |
+
|
| 78 |
+
def evaluate(self, question, gold_answer, pred_answer):
|
| 79 |
+
messages = self.conversation + [
|
| 80 |
+
{"role": "user", "content": f"Question: {question}\nGround Truth Answer: {gold_answer}\nPredicted Answer: {pred_answer}"}
|
| 81 |
+
]
|
| 82 |
+
|
| 83 |
+
try:
|
| 84 |
+
response = self.client.chat.completions.create(
|
| 85 |
+
model="gpt-4", # 或者使用 "gpt-3.5-turbo"
|
| 86 |
+
messages=messages,
|
| 87 |
+
temperature=0,
|
| 88 |
+
max_tokens=1
|
| 89 |
+
)
|
| 90 |
+
score = int(response.choices[0].message.content[0])
|
| 91 |
+
return score if score in [1, 2, 3, 4, 5] else 1
|
| 92 |
+
except Exception as e:
|
| 93 |
+
print(f"Error in GPT evaluation: {e}")
|
| 94 |
+
return 1
|
| 95 |
+
|
| 96 |
+
class DeepSeekCorrectnessEvaluator:
|
| 97 |
+
def __init__(self, api_key):
|
| 98 |
+
api_key = ""
|
| 99 |
+
if api_key == "":
|
| 100 |
+
print("Please set the api key for DeepSeek")
|
| 101 |
+
exit()
|
| 102 |
+
self.client = openai.OpenAI(
|
| 103 |
+
api_key=api_key,
|
| 104 |
+
base_url="https://api.deepseek.com",
|
| 105 |
+
)
|
| 106 |
+
self.conversation = EVALUATOR_PROMPT
|
| 107 |
+
|
| 108 |
+
def evaluate(self, question, gold_answer, pred_answer):
|
| 109 |
+
messages = self.conversation + [
|
| 110 |
+
{"role": "user", "content": f"Question: {question}\nGround Truth Answer: {gold_answer}\nPredicted Answer: {pred_answer}"}
|
| 111 |
+
]
|
| 112 |
+
|
| 113 |
+
try:
|
| 114 |
+
response = self.client.chat.completions.create(
|
| 115 |
+
model="deepseek-chat",
|
| 116 |
+
messages=messages,
|
| 117 |
+
)
|
| 118 |
+
score = int(response.choices[0].message.content[0])
|
| 119 |
+
return score if score in [1, 2, 3, 4, 5] else 1
|
| 120 |
+
except Exception as e:
|
| 121 |
+
print(f"Error in DeepSeek evaluation: {e}")
|
| 122 |
+
return 1
|
| 123 |
+
|
| 124 |
+
def list_user_query(conversation):
|
| 125 |
+
user_query = []
|
| 126 |
+
query_time = []
|
| 127 |
+
for i in range(len(conversation)):
|
| 128 |
+
if conversation[i]['role'].lower() == 'user':
|
| 129 |
+
user_query.append(conversation[i]['content'])
|
| 130 |
+
query_time.append(conversation[i]['time'])
|
| 131 |
+
# Sort user queries and query times by time
|
| 132 |
+
if len(user_query) > 1:
|
| 133 |
+
# Create pairs of (query, time) and sort by time
|
| 134 |
+
query_time_pairs = list(zip(user_query, query_time))
|
| 135 |
+
query_time_pairs.sort(key=lambda x: x[1])
|
| 136 |
+
|
| 137 |
+
# Unpack the sorted pairs back into separate lists
|
| 138 |
+
user_query = [pair[0] for pair in query_time_pairs]
|
| 139 |
+
query_time = [pair[1] for pair in query_time_pairs]
|
| 140 |
+
|
| 141 |
+
return user_query, query_time
|
| 142 |
+
|
| 143 |
+
|
| 144 |
+
def parser_model_output(eval_model, pred_output):
|
| 145 |
+
if eval_model in ['VideollmOnline', 'EWO']: # for context have time information, delete inference time
|
| 146 |
+
if eval_model not in pred_output:
|
| 147 |
+
return [], []
|
| 148 |
+
model_response_list = []
|
| 149 |
+
|
| 150 |
+
pattern = r'\(Video Time = [\d.]+s\)\s*(\w+):\s*(.*?)(?=\s*\(|$)'
|
| 151 |
+
preds = pred_output[eval_model]
|
| 152 |
+
for pred in preds:
|
| 153 |
+
if pred['role'].lower() not in ['assistant', 'user']:
|
| 154 |
+
continue
|
| 155 |
+
match = re.match(pattern, pred['content'])
|
| 156 |
+
if match:
|
| 157 |
+
role = match.group(1)
|
| 158 |
+
content = match.group(2)
|
| 159 |
+
model_response_list.append({'role': role.lower(), 'content': content, 'time': pred['time']})
|
| 160 |
+
|
| 161 |
+
if len(model_response_list) == 0:
|
| 162 |
+
return [], []
|
| 163 |
+
answers = [e for e in model_response_list if e['role'] == 'assistant']
|
| 164 |
+
pred_list = [e['content'] for e in answers]
|
| 165 |
+
pred_time_list = [e['time'] for e in answers]
|
| 166 |
+
else:
|
| 167 |
+
if eval_model not in pred_output:
|
| 168 |
+
return [], []
|
| 169 |
+
model_response_list = []
|
| 170 |
+
preds = pred_output[eval_model]
|
| 171 |
+
for pred in preds:
|
| 172 |
+
if pred['role'].lower() != 'assistant':
|
| 173 |
+
continue
|
| 174 |
+
model_response_list.append({'role': pred['role'].lower(), 'content': pred['content'], 'time': pred['time']})
|
| 175 |
+
|
| 176 |
+
if len(model_response_list) == 0:
|
| 177 |
+
return [], []
|
| 178 |
+
answers = [e for e in model_response_list if e['role'].lower() == 'assistant']
|
| 179 |
+
pred_list = [e['content'] for e in answers]
|
| 180 |
+
pred_time_list = [e['time'] for e in answers]
|
| 181 |
+
|
| 182 |
+
return pred_list, pred_time_list
|
| 183 |
+
|
| 184 |
+
def calculateResponseScore(gold_timespan, pred_time, task_type, anticipation, latency):
|
| 185 |
+
if task_type.strip() in ['Object Localization', 'Object Recognition','Ego Object Localization','Object Function','Attribute Perception',
|
| 186 |
+
'Information Function','Text-Rich Understanding']:
|
| 187 |
+
return (1 - abs(pred_time - (gold_timespan[0])) / (gold_timespan[1] - gold_timespan[0] + latency + anticipation)) * 5
|
| 188 |
+
else:
|
| 189 |
+
return (1 - abs(pred_time - (gold_timespan[0]+gold_timespan[1])/2) / (gold_timespan[1] - gold_timespan[0] + latency + anticipation)) * 5
|
| 190 |
+
|
| 191 |
+
def evalSingleQA(args, f_out, answer_pred, evaluator):
|
| 192 |
+
for video_uid in tqdm(answer_pred.keys()):
|
| 193 |
+
for clip_uid in answer_pred[video_uid].keys():
|
| 194 |
+
for i,qa in enumerate(answer_pred[video_uid][clip_uid]):
|
| 195 |
+
|
| 196 |
+
qa['id'] = i
|
| 197 |
+
qa['video_uid'] = video_uid
|
| 198 |
+
qa['clip_uid'] = clip_uid
|
| 199 |
+
|
| 200 |
+
# parser pred output and gold output
|
| 201 |
+
pred_list, pred_time_list = parser_model_output(args.eval_model, qa)
|
| 202 |
+
gold_list = [e['content'] for e in qa['conversation'] if e['role'].lower() == 'assistant']
|
| 203 |
+
clip_end_time = qa['clip_end_time'] if 'clip_end_time' in qa.keys() else qa['end_time']
|
| 204 |
+
gold_timespan_list = [(ceil_time_by_fps(e['start_time'], 2, 0, clip_end_time),
|
| 205 |
+
ceil_time_by_fps(e['end_time'], 2, 0, clip_end_time)) for e in qa['conversation'] if e['role'].lower() == 'assistant']
|
| 206 |
+
|
| 207 |
+
if 'question' not in qa:
|
| 208 |
+
for e in qa['conversation']:
|
| 209 |
+
if e['role'].lower() == 'user':
|
| 210 |
+
question = e['content']
|
| 211 |
+
break
|
| 212 |
+
else:
|
| 213 |
+
question = qa['question']
|
| 214 |
+
|
| 215 |
+
# construct pred and answer map
|
| 216 |
+
pred_text_to_turn_i = dict()
|
| 217 |
+
for turn_i, text in enumerate(pred_list):
|
| 218 |
+
if text not in pred_text_to_turn_i:
|
| 219 |
+
pred_text_to_turn_i[text] = list()
|
| 220 |
+
pred_text_to_turn_i[text].append(turn_i)
|
| 221 |
+
|
| 222 |
+
gold_text_to_turn_i = dict()
|
| 223 |
+
for turn_i, text in enumerate(gold_list):
|
| 224 |
+
if text not in gold_text_to_turn_i:
|
| 225 |
+
gold_text_to_turn_i[text] = list()
|
| 226 |
+
gold_text_to_turn_i[text].append(turn_i)
|
| 227 |
+
|
| 228 |
+
|
| 229 |
+
score_matrix = np.zeros((len(gold_list), len(pred_list)))
|
| 230 |
+
response_score_matrix = np.zeros((len(gold_list), len(pred_list)))
|
| 231 |
+
|
| 232 |
+
for gold_content, gold_turn_ids in gold_text_to_turn_i.items():
|
| 233 |
+
for pred_content, pred_turn_ids in pred_text_to_turn_i.items():
|
| 234 |
+
# we only need to evaluate the pred answer that is in the gold span to for the in-span metric
|
| 235 |
+
gold_timespan = [gold_timespan_list[i] for i in gold_turn_ids]
|
| 236 |
+
pred_time = [pred_time_list[i] for i in pred_turn_ids]
|
| 237 |
+
# the pred answer with time -1 can pair with any other span
|
| 238 |
+
pred_time_in_gold_timespan_list = [(time == -1 or span[0]-args.anticipation <= time <= span[1]+args.latency) for time in pred_time for span in gold_timespan]
|
| 239 |
+
if not any(pred_time_in_gold_timespan_list):
|
| 240 |
+
continue
|
| 241 |
+
|
| 242 |
+
if args.eval_model in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 243 |
+
score = 0
|
| 244 |
+
else:
|
| 245 |
+
score = evaluator.evaluate(question, gold_content, pred_content)
|
| 246 |
+
|
| 247 |
+
row_indices, col_indices = np.meshgrid(gold_turn_ids, pred_turn_ids)
|
| 248 |
+
|
| 249 |
+
for i, (row, col) in enumerate(zip(row_indices.flatten(), col_indices.flatten())):
|
| 250 |
+
if pred_time_in_gold_timespan_list[i]:
|
| 251 |
+
text_score = score
|
| 252 |
+
reponse_score = calculateResponseScore(gold_timespan_list[row], pred_time_list[col], qa['Task Type'], args.anticipation, args.latency)
|
| 253 |
+
score_matrix[row, col] = text_score
|
| 254 |
+
response_score_matrix[row, col] = reponse_score
|
| 255 |
+
|
| 256 |
+
|
| 257 |
+
qa['evaluator_output_text'] = score_matrix.tolist()
|
| 258 |
+
qa['evaluator_output_reponse'] = response_score_matrix.tolist()
|
| 259 |
+
with open(f_out, 'a') as f:
|
| 260 |
+
f.write(json.dumps(qa,indent=4) + '\n')
|
| 261 |
+
|
| 262 |
+
|
| 263 |
+
def evalMultiQA(args, f_out, answer_pred, evaluator):
|
| 264 |
+
for video_uid in tqdm(answer_pred.keys()):
|
| 265 |
+
for clip_uid in answer_pred[video_uid].keys():
|
| 266 |
+
for i,qa in enumerate(answer_pred[video_uid][clip_uid]):
|
| 267 |
+
|
| 268 |
+
qa['id'] = i
|
| 269 |
+
qa['video_uid'] = video_uid
|
| 270 |
+
qa['clip_uid'] = clip_uid
|
| 271 |
+
|
| 272 |
+
for conv in qa['conversation']:
|
| 273 |
+
if 'time' not in conv.keys():
|
| 274 |
+
conv['time'] = conv['start_time'] + (conv['end_time'] - conv['start_time']) / 2
|
| 275 |
+
qa['conversation'] = sorted(qa['conversation'], key=lambda x: x['time'])
|
| 276 |
+
|
| 277 |
+
questions, query_times = list_user_query(qa['conversation'])
|
| 278 |
+
start_time = qa['clip_start_time'] if 'clip_start_time' in qa.keys() else qa['start_time']
|
| 279 |
+
end_time = qa['clip_end_time'] if 'clip_end_time' in qa.keys() else qa['end_time']
|
| 280 |
+
max_time = qa['conversation'][-1]['end_time']
|
| 281 |
+
|
| 282 |
+
start_time = min([start_time]+query_times)
|
| 283 |
+
end_time = min([end_time,max_time])
|
| 284 |
+
|
| 285 |
+
start_times = []
|
| 286 |
+
for question, query_time in zip(questions, query_times):
|
| 287 |
+
start_times.append(query_time)
|
| 288 |
+
end_times = start_times[1:] + [end_time]
|
| 289 |
+
for question_idx, (question, query_time, start_time, end_time) in enumerate(zip(questions, query_times, start_times, end_times)):
|
| 290 |
+
# parser pred output and gold output
|
| 291 |
+
pred_list, pred_time_list = parser_model_output(args.eval_model, qa)
|
| 292 |
+
|
| 293 |
+
# HACK: add in-span pred
|
| 294 |
+
in_span_pred_list, in_span_pred_time_list = [], []
|
| 295 |
+
for pred, pred_time in zip(pred_list, pred_time_list):
|
| 296 |
+
if start_time <= pred_time <= end_time:
|
| 297 |
+
in_span_pred_list.append(pred)
|
| 298 |
+
in_span_pred_time_list.append(pred_time)
|
| 299 |
+
pred_list, pred_time_list = in_span_pred_list, in_span_pred_time_list
|
| 300 |
+
|
| 301 |
+
gold_list = [e['content'] for e in qa['conversation'] if e['role'].lower() == 'assistant']
|
| 302 |
+
clip_end_time = qa['clip_end_time'] if 'clip_end_time' in qa.keys() else qa['end_time']
|
| 303 |
+
gold_timespan_list = [(ceil_time_by_fps(e['start_time'], 2, 0, clip_end_time),
|
| 304 |
+
ceil_time_by_fps(e['end_time'], 2, 0, clip_end_time)) for e in qa['conversation'] if e['role'].lower() == 'assistant']
|
| 305 |
+
|
| 306 |
+
|
| 307 |
+
# HACK: add in-span gold
|
| 308 |
+
in_span_gold_list, in_span_gold_timespan_list = [], []
|
| 309 |
+
for gold, gold_time in zip(gold_list, gold_timespan_list):
|
| 310 |
+
if start_time <= gold_time[0] <= end_time and start_time <= gold_time[1] <= end_time:
|
| 311 |
+
in_span_gold_list.append(gold)
|
| 312 |
+
in_span_gold_timespan_list.append(gold_time)
|
| 313 |
+
gold_list, gold_timespan_list = in_span_gold_list, in_span_gold_timespan_list
|
| 314 |
+
|
| 315 |
+
|
| 316 |
+
|
| 317 |
+
# construct pred and answer map
|
| 318 |
+
pred_text_to_turn_i = dict()
|
| 319 |
+
for turn_i, text in enumerate(pred_list):
|
| 320 |
+
if text not in pred_text_to_turn_i:
|
| 321 |
+
pred_text_to_turn_i[text] = list()
|
| 322 |
+
pred_text_to_turn_i[text].append(turn_i)
|
| 323 |
+
|
| 324 |
+
gold_text_to_turn_i = dict()
|
| 325 |
+
for turn_i, text in enumerate(gold_list):
|
| 326 |
+
if text not in gold_text_to_turn_i:
|
| 327 |
+
gold_text_to_turn_i[text] = list()
|
| 328 |
+
gold_text_to_turn_i[text].append(turn_i)
|
| 329 |
+
|
| 330 |
+
score_matrix = np.zeros((len(gold_list), len(pred_list)))
|
| 331 |
+
response_score_matrix = np.zeros((len(gold_list), len(pred_list)))
|
| 332 |
+
|
| 333 |
+
for gold_content, gold_turn_ids in gold_text_to_turn_i.items():
|
| 334 |
+
for pred_content, pred_turn_ids in pred_text_to_turn_i.items():
|
| 335 |
+
# we only need to evaluate the pred answer that is in the gold span to for the in-span metric
|
| 336 |
+
gold_timespan = [gold_timespan_list[i] for i in gold_turn_ids]
|
| 337 |
+
pred_time = [pred_time_list[i] for i in pred_turn_ids]
|
| 338 |
+
# the pred answer with time -1 can pair with any other span
|
| 339 |
+
pred_time_in_gold_timespan_list = [(time == -1 or span[0]-args.anticipation <= time <= span[1]+args.latency) for time in pred_time for span in gold_timespan]
|
| 340 |
+
if not any(pred_time_in_gold_timespan_list):
|
| 341 |
+
continue
|
| 342 |
+
|
| 343 |
+
if args.eval_model in ['EgoVLP', 'CLIP', 'Lavila']:
|
| 344 |
+
score = 0
|
| 345 |
+
else:
|
| 346 |
+
score = evaluator.evaluate(question, gold_content, pred_content)
|
| 347 |
+
|
| 348 |
+
row_indices, col_indices = np.meshgrid(gold_turn_ids, pred_turn_ids)
|
| 349 |
+
|
| 350 |
+
for i, (row, col) in enumerate(zip(row_indices.flatten(), col_indices.flatten())):
|
| 351 |
+
if pred_time_in_gold_timespan_list[i]:
|
| 352 |
+
text_score = score
|
| 353 |
+
reponse_score = calculateResponseScore(gold_timespan_list[row], pred_time_list[col], qa['Task Type'], args.anticipation, args.latency)
|
| 354 |
+
score_matrix[row, col] = text_score
|
| 355 |
+
response_score_matrix[row, col] = reponse_score
|
| 356 |
+
|
| 357 |
+
|
| 358 |
+
qa['evaluator_output_text'] = score_matrix.tolist()
|
| 359 |
+
qa['evaluator_output_reponse'] = response_score_matrix.tolist()
|
| 360 |
+
with open(f_out, 'a') as f:
|
| 361 |
+
f.write(json.dumps(qa,indent=4) + '\n')
|
| 362 |
+
|
| 363 |
+
|
| 364 |
+
def calQuestionTotal(answer_pred):
|
| 365 |
+
c = 0
|
| 366 |
+
for k,v in answer_pred.items():
|
| 367 |
+
for kk,vv in v.items():
|
| 368 |
+
for qa in vv:
|
| 369 |
+
if args.eval_model in qa:
|
| 370 |
+
c+=1
|
| 371 |
+
print('total question:', c)
|
| 372 |
+
|
| 373 |
+
def main_worker(rank, world_size, args):
|
| 374 |
+
# 设置当前进程使用的 GPU
|
| 375 |
+
torch.cuda.set_device(rank)
|
| 376 |
+
args.device = f"cuda:{rank}"
|
| 377 |
+
|
| 378 |
+
# 初始化分布式环境（此处采用 NCCL 后端，适用于 GPU 之间的通信）
|
| 379 |
+
dist.init_process_group(backend="nccl", init_method="env://", rank=rank, world_size=world_size)
|
| 380 |
+
|
| 381 |
+
# 加载数据
|
| 382 |
+
if args.concat:
|
| 383 |
+
parent_dir = os.path.dirname(args.pred_file)
|
| 384 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(args.pred_file.split('/')[-1])]
|
| 385 |
+
data = {}
|
| 386 |
+
for eval_file in eval_files:
|
| 387 |
+
data.update(json.load(open(eval_file)))
|
| 388 |
+
else:
|
| 389 |
+
print(args.pred_file)
|
| 390 |
+
data = json.load(open(args.pred_file))
|
| 391 |
+
|
| 392 |
+
# 将数据分片（此处假设 data 为列表；如果是字典，需要根据实际情况修改分片方式）
|
| 393 |
+
if isinstance(data, list):
|
| 394 |
+
local_data = data[rank::world_size]
|
| 395 |
+
elif isinstance(data, dict):
|
| 396 |
+
keys = list(data.keys())
|
| 397 |
+
local_keys = keys[rank::world_size]
|
| 398 |
+
local_data = {k: data[k] for k in local_keys}
|
| 399 |
+
else:
|
| 400 |
+
local_data = data # 未知结构则不分片
|
| 401 |
+
|
| 402 |
+
output_file = args.pred_file.replace('.json', f'evaluator_{args.evaluator_llm}_{args.anticipation}_{args.latency}.json')
|
| 403 |
+
local_output_file = f"{output_file}.part{rank}"
|
| 404 |
+
# Create an empty local output file
|
| 405 |
+
with open(local_output_file, 'w') as f:
|
| 406 |
+
f.write('') # Write empty string to create the file
|
| 407 |
+
f_out = local_output_file
|
| 408 |
+
print(f_out)
|
| 409 |
+
|
| 410 |
+
# init
|
| 411 |
+
if args.evaluator_llm == 'gpt':
|
| 412 |
+
evaluator = GPTCorrectnessEvaluator('sk-proj-1234567890')
|
| 413 |
+
elif args.evaluator_llm == 'llama':
|
| 414 |
+
evaluator = CorrectnessEvaluator('meta-llama/Meta-Llama-3-8B-Instruct')
|
| 415 |
+
elif args.evaluator_llm == 'deepseek':
|
| 416 |
+
evaluator = DeepSeekCorrectnessEvaluator('sk-43a08cfb3ae64b6288ee67db8009c8ca')
|
| 417 |
+
else:
|
| 418 |
+
raise ValueError(f'evaluator_llm {args.evaluator_llm} not supported')
|
| 419 |
+
|
| 420 |
+
print('start eval')
|
| 421 |
+
# eval
|
| 422 |
+
if args.qa_type == 'SingleQA':
|
| 423 |
+
evalSingleQA(args, f_out, local_data, evaluator)
|
| 424 |
+
elif args.qa_type == 'MultiQA':
|
| 425 |
+
evalMultiQA(args, f_out, local_data, evaluator)
|
| 426 |
+
else:
|
| 427 |
+
raise ValueError(f'qa_type {args.qa_type} not supported')
|
| 428 |
+
|
| 429 |
+
|
| 430 |
+
def main(args):
|
| 431 |
+
print(args.pred_file, args.concat)
|
| 432 |
+
if args.concat:
|
| 433 |
+
parent_dir = os.path.dirname(args.pred_file)
|
| 434 |
+
eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(args.pred_file.split('/')[-1])]
|
| 435 |
+
data = {}
|
| 436 |
+
for eval_file in eval_files:
|
| 437 |
+
data.update(json.load(open(eval_file)))
|
| 438 |
+
else:
|
| 439 |
+
data = json.load(open(args.pred_file))
|
| 440 |
+
calQuestionTotal(data)
|
| 441 |
+
|
| 442 |
+
if torch.cuda.device_count() < 2:
|
| 443 |
+
# init
|
| 444 |
+
if args.evaluator_llm == 'gpt':
|
| 445 |
+
evaluator = GPTCorrectnessEvaluator('sk-proj-1234567890')
|
| 446 |
+
elif args.evaluator_llm == 'llama':
|
| 447 |
+
evaluator = CorrectnessEvaluator('meta-llama/Meta-Llama-3-8B-Instruct')
|
| 448 |
+
elif args.evaluator_llm == 'deepseek':
|
| 449 |
+
evaluator = DeepSeekCorrectnessEvaluator('sk-43a08cfb3ae64b6288ee67db8009c8ca')
|
| 450 |
+
else:
|
| 451 |
+
raise ValueError(f'evaluator_llm {args.evaluator_llm} not supported')
|
| 452 |
+
|
| 453 |
+
output_file = args.pred_file.replace('.json', f'_evaluator_{args.evaluator_llm}_{args.anticipation}_{args.latency}.json')
|
| 454 |
+
with open(output_file, 'w') as f:
|
| 455 |
+
f.write('')
|
| 456 |
+
f_out = output_file
|
| 457 |
+
|
| 458 |
+
# eval
|
| 459 |
+
if args.qa_type == 'SingleQA':
|
| 460 |
+
evalSingleQA(args, f_out, data, evaluator)
|
| 461 |
+
elif args.qa_type == 'MultiQA':
|
| 462 |
+
evalMultiQA(args, f_out, data, evaluator)
|
| 463 |
+
else:
|
| 464 |
+
raise ValueError(f'qa_type {args.qa_type} not supported')
|
| 465 |
+
|
| 466 |
+
else:
|
| 467 |
+
os.environ['MASTER_ADDR'] = '127.0.0.1'
|
| 468 |
+
os.environ['MASTER_PORT'] = str(args.master_port)
|
| 469 |
+
world_size = torch.cuda.device_count()
|
| 470 |
+
mp.spawn(main_worker, args=(world_size, args), nprocs=world_size, join=True)
|
| 471 |
+
|
| 472 |
+
|
| 473 |
+
if __name__ == '__main__':
|
| 474 |
+
parser = argparse.ArgumentParser()
|
| 475 |
+
parser.add_argument('--evaluator_llm', type=str, default='gpt-4')
|
| 476 |
+
parser.add_argument('--eval_model', type=str, default='VideollmOnline')
|
| 477 |
+
parser.add_argument('--pred_file', type=str, default='/root/videollm-online/data/estp_dataset/tmp_predict_VideollmOnline_v2.json')
|
| 478 |
+
parser.add_argument('--concat', type=bool, default=False)
|
| 479 |
+
parser.add_argument('--qa_type', type=str, default='SingleQA')
|
| 480 |
+
parser.add_argument('--anticipation', type=int, default=0.0)
|
| 481 |
+
parser.add_argument('--latency', type=int, default=0.0)
|
| 482 |
+
parser.add_argument('--master_port', type=int, default=29501)
|
| 483 |
+
args = parser.parse_args()
|
| 484 |
+
main(args)
|
ESTP-Bench/estp_dataset/benchmark/merge_prediction_result.ipynb
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cells": [
|
| 3 |
+
{
|
| 4 |
+
"cell_type": "markdown",
|
| 5 |
+
"metadata": {},
|
| 6 |
+
"source": [
|
| 7 |
+
"# Concat generated prediction results from multiple files"
|
| 8 |
+
]
|
| 9 |
+
},
|
| 10 |
+
{
|
| 11 |
+
"cell_type": "code",
|
| 12 |
+
"execution_count": null,
|
| 13 |
+
"metadata": {},
|
| 14 |
+
"outputs": [],
|
| 15 |
+
"source": [
|
| 16 |
+
"import os\n",
|
| 17 |
+
"import json\n",
|
| 18 |
+
"\n",
|
| 19 |
+
"pred_file = 'estp_bench_sq_EWO_frame_by_frame_fusion_dinov2_first_5_cases.json'\n",
|
| 20 |
+
"parent_dir = os.path.dirname(pred_file)\n",
|
| 21 |
+
"eval_files = [os.path.join(parent_dir, f) for f in os.listdir(parent_dir) if f.startswith(pred_file.split('/')[-1])]\n",
|
| 22 |
+
"data = {}\n",
|
| 23 |
+
"for eval_file in eval_files:\n",
|
| 24 |
+
" data.update(json.load(open(eval_file)))\n",
|
| 25 |
+
"\n",
|
| 26 |
+
"with open(pred_file, 'w') as f:\n",
|
| 27 |
+
" json.dump(data, f, indent=4)"
|
| 28 |
+
]
|
| 29 |
+
}
|
| 30 |
+
],
|
| 31 |
+
"metadata": {
|
| 32 |
+
"kernelspec": {
|
| 33 |
+
"display_name": "videollm",
|
| 34 |
+
"language": "python",
|
| 35 |
+
"name": "python3"
|
| 36 |
+
},
|
| 37 |
+
"language_info": {
|
| 38 |
+
"name": "python",
|
| 39 |
+
"version": "3.10.14"
|
| 40 |
+
}
|
| 41 |
+
},
|
| 42 |
+
"nbformat": 4,
|
| 43 |
+
"nbformat_minor": 2
|
| 44 |
+
}
|
ESTP-Bench/estp_dataset/cqa_anno.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/dataset.py
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part0
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part1
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part2
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part3
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part4
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part5
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part6
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/CLIP_streaming_v2evaluator_deepseek_1_2.json.part7
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part0
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part1
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part2
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part3
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part4
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part5
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part6
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/EgoVLP_streaming_v2evaluator_deepseek_1_2.json.part7
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part0
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175.json.part1
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part0
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part1
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part2
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part3
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part4
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part5
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part6
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
ESTP-Bench/estp_dataset/estpCqa_baseline/InternVLV28_fbf_0.175evaluator_deepseek_1_2.json.part7
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|