Object-centric structural video control on BEHAVE
95.8ContainmentVACE
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VACEInput=2D Boxes, Finetuned=true2026.06 | 95.8 | 48.2 | 23.8 | 41.2 | 14.9 | 49.7 | 69.3 | |
| LCVInput=Rendered 3D Boxes, Finetuned=false2026.06 | 88.6 | 95.6 | 6.2 | 20.7 | 5.8 | 90.2 | 75 | |
| VACEInput=2D Flow, Finetuned=true2026.06 | 63.8 | 84.7 | 9.8 | 31.8 | 7.6 | 78.5 | 75.8 | |
| VACEInput=2D Flow, Finetuned=false2026.06 | 57.2 | 86.3 | 12.1 | 34.1 | 8.9 | 74.8 | 75.1 | |
| Control-freeInput=GT First + Last, Finetuned=false2026.06 | 42.3 | 46.9 | 61.1 | 49 | 54.8 | 48.6 | 76.2 |