Correlation to Human Preference on Human Preference Annotations Physics Scenarios (test)
39.4Correlation: Ball CollisionQwen2.5 VL
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5 VL2025.10 | 39.4 | 59 | -25.4 | -19.4 | 14.2 | 49.5 | -15.2 | 29.1 | -22.2 | 26.9 | 45.2 | -33.9 | 33.3 | |
| LikePhys2025.10 | 34.4 | 42.6 | 24.3 | 15 | 55.9 | 15.4 | 0 | 51 | -8.7 | 16.8 | 36.4 | 12.1 | 44.4 | |
| VideoPhy2025.10 | 20 | -11 | 26.7 | -13.8 | -20.9 | 57.6 | 15.3 | 26.3 | -6.4 | 40.8 | 47.5 | -9.6 | 38.9 | |
| VideoPhy22025.10 | -19.4 | -32.2 | -6.8 | -15.9 | -3.1 | 51 | 10.5 | -20 | -41 | 6.8 | 70.4 | -33.9 | -8.5 |