Camera-view-conditioned Video Generation on RealEstate10K long-term (256-Frame)
237FVDGeometry Forcing + REPA
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Geometry Forcing + REPAFrames=2562025.07 | 237 | 0.51 | 0.37 | 12.1 | 0.3264 | 236 | |
| Geometry ForcingFrames=2562025.07 | 243 | 0.51 | 0.38 | 11.87 | 0.3337 | 272 | |
| REPAFrames=2562025.07 | 297 | 0.54 | 0.36 | 11.51 | 0.3337 | 315 | |
| DFoTFrames=2562025.07 | 364 | 0.55 | 0.36 | 11.4 | 0.3575 | 297 | |
| VideoREPAFrames=2562025.07 | 455 | 0.56 | 0.35 | 11.5 | 0.3823 | 190 | |
| CosmosFrames=256, Conditioning=first frame only2025.07 | 934 | 0.68 | 0.2 | 10.25 | — | — |