Viewpoint Grounding on ScanRefer
38.6R@1UniScene3D
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UniScene3DInput=colored pointmaps (CPM), Zero-shot=true2026.04 | 38.6 | 72.5 | 85.7 | |
| SigLIP2Input=multi-view images (I), Zero-shot=true2026.04 | 22.2 | 61.1 | 79.5 | |
| DFNInput=multi-view images (I), Zero-shot=true2026.04 | 20.9 | 59.2 | 77.6 | |
| FG-CLIPInput=multi-view images (I), Zero-shot=true2026.04 | 20.2 | 58.5 | 77.6 | |
| POMA-3DInput=pointmaps (PM), Zero-shot=true2026.04 | 16.4 | 50.6 | 71.2 | |
| Uni3D-gInput=point clouds (PC), Zero-shot=true2026.04 | 4.2 | 19 | 35.5 |