Long-horizon Video Prediction on EgoDex (65 frames)
22.75PSNRBase + DINOv3
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Base + DINOv3Rollout Length=65 frames, Base Model=Cosmos-Predict2.5-2B2026.06 | 22.75 | 76 | 0.31 | 134.12 | 499.44 | 26.45 | 41.73 | |
| DexAC-WM (Base + DINOv3 + DexAC)Rollout Length=65 frames, Base Model=Cosmos-Predict2.5-2B, Components=DINOv3 + DexAC2026.06 | 22.69 | 76 | 0.31 | 133.02 | 465.31 | 27.42 | 50.67 | |
| Base + DexACRollout Length=65 frames, Base Model=Cosmos-Predict2.5-2B2026.06 | 22.48 | 75 | 0.32 | 134.28 | 484.34 | 27.22 | 50.7 | |
| Cosmos-Predict2.5-2B (Base)Rollout Length=65 frames2026.06 | 21.89 | 74 | 0.34 | 158.7 | 524.96 | 24.45 | 47.93 |