3D Instance Segmentation on ScanNet V2
81.5AP@50%SegDINO3D
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SegDINO3DInput Modality=P & I & D & C2026.03 | 81.5 | 64 | 88.9 | |
| SGIFormer-LInput Modality=P2026.03 | 81.2 | 61 | 88.9 | |
| Relation3DInput Modality=P2026.03 | 80.2 | 62.5 | 87 | |
| OneFormer3DInput Modality=P2026.03 | 78.1 | 59.3 | 86.4 | |
| MAFTInput Modality=P2026.03 | 76.5 | 59.9 | — | |
| QueryFormerInput Modality=P2026.03 | 74.2 | 56.5 | 83.3 | |
| SPFormerInput Modality=P2026.03 | 73.9 | 56.3 | 82.9 | |
| Mask3DInput Modality=P2026.03 | 73.7 | 55.2 | 85.3 | |
| OursPoints per Scene=90 (test), Points per Objects=3, Objects per Scene=≈ 302022.04 | 73.4 | — | — | |
| SegVGGTInput Modality=I2026.03 | 71.7 | 50.4 | 87 | |
| ODINInput Modality=I & D & C2026.03 | 71 | 50 | 83.6 | |
| SoftGroup++ & TAMMpre-training=TAMM, backbone=Point-BERT2024.02 | 68 | 46.1 | 79 | |
| SoftGroup++ & OpenShapepre-training=OpenShape, backbone=Point-BERT2024.02 | 67.4 | 45.7 | 78.7 | |
| SoftGroup++implementation=Reproduced, backbone=SparseConv2024.02 | 67 | 45.5 | 78.7 | |
| OursPoints per Scene=60 (test), Points per Objects=2, Objects per Scene=≈ 302022.04 | 60.8 | — | — | |
| CSCPoints per Scene=200 (train)2022.04 | 48.9 | — | — | |
| IGGT2026.03 | 11.2 | 2.8 | 28.7 | |
| OneFormer3DInput Modality=I, Reconstruction=3D reconstruction with VGGT [52]2026.03 | 10.2 | 5.4 | 17.4 | |
| FAST3DIS2026.03 | 9.6 | 3.8 | 31.6 | |
| SAM3D2026.03 | 8 | 2.4 | 23.3 | |
| Segment3D2026.03 | 7.2 | 2.5 | 17.2 | |
| PanSt3R2026.03 | 3.5 | 0.8 | 22.5 | |
| Felzenszwalb2026.03 | 2.2 | 0.8 | 10 | |
| HDBSCAN2026.03 | 0.6 | 0.2 | 5.6 |