Video Physical Commonsense Evaluation on VideoPhy-2
0.76Spearman PCVIDEOPHY–2–AUTOEVAL
Evaluation Results
| Method | Links | |
|---|---|---|
| VIDEOPHY–2–AUTOEVALfine-tuned=true2026.02 | 0.76 | |
| 3DSPA (ours)fine-tuned=false2026.02 | 0.74 | |
| 3DSPA (no DINO)fine-tuned=false, DINO features=false2026.02 | 0.5 | |
| VideoCon–Physicsfine-tuned=true2026.02 | 0.48 | |
| 3DSPA (no 3D)fine-tuned=false, 3D structure=false2026.02 | 0.4 | |
| 3DSPA (no 3D, no DINO)fine-tuned=false, 3D structure=false, DINO features=false2026.02 | 0.19 | |
| VideoScorefine-tuned=true2026.02 | 0.17 | |
| VideoConfine-tuned=true2026.02 | 0.13 | |
| VideoLlavafine-tuned=true2026.02 | 0.08 |