Cross-Modal Scene Retrieval (Point Cloud to Text) on ScanNet
57.73Scene Category Recall@1CrossOver
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| CrossOverDirection=P → R2025.02 | 57.73 | 79.04 | 85.57 | 5 | 20 | 46 | 26.79 | 56.67 | 68.63 | |
| Inst. Baseline (Ours)Direction=P → R2025.02 | 28.87 | 50.86 | 66.67 | 5 | 13 | 23 | 29.41 | 50.84 | 65.65 | |
| PointBindDirection=P → R2025.02 | 18.21 | 26.46 | 31.96 | 1 | 2 | 6 | 18.25 | 40.05 | 54.84 | |
| ULIP-2Direction=P → R2025.02 | 11.34 | 15.12 | 23.27 | 1 | 2 | 4 | 18.12 | 41.15 | 54.93 |