Object-centric structural video control on HO-3D
97.9ContainVACE
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VACEInput=2D Boxes, Finetuned=true2026.06 | 97.9 | 56.8 | 12.6 | 18.1 | 9.7 | 55.1 | 72.4 | |
| LCVInput=Rendered 3D Boxes, Finetuned=false2026.06 | 91.3 | 97.4 | 4.5 | 12.2 | 3.9 | 94.1 | 72.9 | |
| VACEInput=2D Flow, Finetuned=true2026.06 | 69.1 | 88.2 | 7.1 | 19.2 | 5.4 | 84.2 | 73.6 | |
| VACEInput=2D Flow, Finetuned=false2026.06 | 63.4 | 89.7 | 8.3 | 21.4 | 6.3 | 81.5 | 73.3 | |
| Control-freeInput=GT First + Last, Finetuned=false2026.06 | 46.8 | 52.1 | 44 | 36.2 | 38.5 | 53.4 | 76.4 |