GeoFix: Geometry Fixing for 4D Consistent Generation
GeoFix targets 4D video geometry with one calibrated depth-consistency reward.
The paper says video diffusion models used for 4D generation often produce geometry that does not stay consistent across views. GeoFix measures cross-view depth agreement in static scene regions, aiming at structure without directly constraining appearance or valid object motion. The authors also correct for a failure mode where smaller camera movement can look like better geometry because view overlap is higher. They report a 39% gain in WorldScore 3D consistency and a 4% rise in VBench Quality Score. HF Daily Papers' note
The paper says video diffusion models used for 4D generation often produce geometry that does not stay consistent across views. GeoFix measures cross-view depth agreement in static scene regions, aiming at structure without directly constraining appearance or valid object motion. The authors also correct for a failure mode where smaller camera movement can look like better geometry because view overlap is higher. They report a 39% gain in WorldScore 3D consistency and a 4% rise in VBench Quality Score. HF Daily Papers' note
score 5