Open-set Segmentation on Replica
84.5mIoUOurs (SAM1.0)
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Ours (SAM1.0)Segmentation Model=SAM 1.02025.12 | 84.5 | — | 83.2 | 76.1 | 87.3 | 83.7 | 88.2 | 85.1 | 84.7 | 87.3 | |
| Gaussian Grouping2025.12 | 69 | — | 61.3 | 63.1 | 73.2 | 74.4 | 68.3 | 76.3 | 66.4 | 68.6 | |
| Feature 3DGS2025.12 | 57.1 | — | 50.6 | 56.9 | 52.7 | 63.3 | 59.8 | 61.3 | 52.9 | 59.3 | |
| FeatureSLAMBackbone=3DGS+GSAM2, Training Time=5.4 min2026.01 | 46.3 | 55.3 | — | — | — | — | — | — | — | — | |
| Feature3DGSBackbone=3DGS+LSeg, Training Time=150 min2026.01 | 39.9 | 52.1 | — | — | — | — | — | — | — | — | |
| LERFBackbone=NeRF+CLIP, Training Time=45 min2026.01 | 31.2 | 60.7 | — | — | — | — | — | — | — | — | |
| LangSplatBackbone=3DGS+CLIP, Training Time=180 min2026.01 | 24.7 | 42 | — | — | — | — | — | — | — | — | |
| OVO-G.-SLAMBackbone=3DGS+CLIP, Training Time=138 min2026.01 | 22.3 | 41.1 | — | — | — | — | — | — | — | — |