curated dataset
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
800-hour curated dataset (test)
100Precision
14
Curated Dataset standard (train-test)
100TPR
11
curated dataset (test)
0BAD BOT Rate
11
Curated dataset 34 examples
69.4Physical Realism
8
Curated dataset of 10 scenes (test)
93.26Camera Control
8
Curated dataset of 10 scenes 1.0 (test)
93.5Physics Plausibility
7
Curated dataset 90-frames
95.99CLIP-F
7
curated dataset
58.3Negative Accuracy
6
Curated dataset 100 image pairs
4.173DeQA
6
Curated dataset 100 image pairs
0.293CLIP-T Score
6
Curated dataset 100 image pairs
88.32CLIP-I
6
Curated dataset 8-frames
95.95CLIP-F
6
Curated dataset 36-frames
9,318CLIP-F
5
Curated dataset of 30 images (test)
89.6Action Following
3