Egocentric latent state prediction on Ego-Exo4D Cooking
0.062L2 Error (2s)Naive EgoExo-WM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Naive EgoExo-WMDescription=Naive use of raw exocentric video2026.05 | 0.062 | 0.052 | 36.8 | 44.3 | |
| EgoExo-WMDescription=Converted exocentric data2026.05 | 0.062 | 0.052 | 46 | 51.5 | |
| Ego-WMDescription=Egocentric-only training2026.05 | 0.063 | 0.053 | 46 | 45.9 | |
| EgoControl*Resolution=224x2242026.05 | 0.09 | 0.078 | 22.3 | 37.8 | |
| PEVA-XXLContext Frames=15, Resolution=224x2242026.05 | 0.102 | 0.095 | 19.7 | 30.3 | |
| PEVA-LContext Frames=7, Resolution=224x2242026.05 | 0.105 | 0.097 | 21 | 29.8 | |
| PEVA-XLContext Frames=15, Resolution=224x2242026.05 | 0.105 | 0.097 | 21 | 30.1 |