3D Semantic Segmentation on ScanNet200 (val)
37.5mIoU (All Classes)PPT (joint-training)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PPT (joint-training)Backbone=PTv3, Training Data=ScanNet + ScanNet200 + ScanNet++ + Structure3D + ARKit LabelMaker2024.10 | 37.5 | 58.8 | 33.3 | 20.4 | — | |
| PTv3 (fine-tune)Backbone=PTv3, Training Data=ARKit LabelMaker -> ScanNet2002024.10 | 37 | 57 | 32.6 | 21.7 | — | |
| PTv3 (fine-tune)Backbone=PTv3, Training Data=ARKit LabelMaker SN200 -> ScanNet2002024.10 | 36.4 | 56.7 | 31.2 | 21.6 | — | |
| PPT (ablation)Backbone=PTv3, Training Data=ScanNet + ScanNet200 + Structure3D -> ScanNet2002024.10 | 36.3 | 56.6 | 31.7 | 20.7 | — | |
| PPTBackbone=PTv3, Training Data=ScanNet200 + S3DIS + Structure3D -> ScanNet2002024.10 | 36 | — | — | — | — | |
| PTv3 (vanilla)Backbone=PTv3, Training Data=ScanNet2002024.10 | 35.2 | 56.5 | 30.1 | 19.3 | — | |
| MinkUNet (co-training)Backbone=MinkUNet, Training Data=ARKit LabelMaker SN200 + ScanNet2002024.10 | 28.2 | 52.5 | 22.4 | 9.8 | — | |
| MinkUNet (fine-tune)Backbone=MinkUNet, Training Data=ARKit LabelMaker SN200 -> ScanNet2002024.10 | 28.1 | 51.6 | 22.3 | 10.6 | — | |
| MinkUNet (vanilla)Backbone=MinkUNet, Training Data=ScanNet2002024.10 | 27.2 | 49.8 | 20.8 | 11.2 | — | |
| PTv3Regime=Supervised reference, Setting=Closed-set, supervised, Input=Depth-only geometry2026.07 | 20.2 | — | — | — | 28.2 | |
| 3D-AVSUnknown Vocabulary=True, Label Set=I+L2024.06 | 14.6 | — | — | — | — | |
| Diff2SceneUnknown Vocabulary=False, Label Set=Official2024.06 | 14.2 | — | — | — | — | |
| Mosaic3D + UTTORegime=Non-private reference, Setting=Open-voc., training-free, Input=RGB-D2026.07 | 13.3 | — | — | — | 26 | |
| Mosaic3D + UTTOSetting=Non-private, Input=RGB-D2026.07 | 13.3 | — | — | — | 26 | |
| Mosaic3DSetting=Non-private, Input=RGB-D2026.07 | 13.1 | — | — | — | 25.3 | |
| OpenSceneUnknown Vocabulary=False, Label Set=Official2024.06 | 11.7 | — | — | — | — | |
| OpenMask3DUnknown Vocabulary=False, Label Set=Official2024.06 | 10.3 | — | — | — | — | |
| ConceptFusionUnknown Vocabulary=False, Label Set=Official2024.06 | 8.8 | — | — | — | — | |
| Mosaic3D-DepthOnly + UTTORegime=Ours, Setting=Open-voc., training-free, Input=Depth-only geometry2026.07 | 6.1 | — | — | — | 11.4 | |
| Mosaic3D-DepthOnly + UTTO (Ours)Setting=Privacy-preserving, Input=Depth-only geometry2026.07 | 6.1 | — | — | — | 11.4 | |
| OpenScene3D + UTTORegime=Ours, Setting=Open-voc., training-free, Input=Depth-only geometry2026.07 | 5.9 | — | — | — | 10.2 | |
| OpenScene3D + UTTO (Ours)Setting=Privacy-preserving, Input=Depth-only geometry2026.07 | 5.9 | — | — | — | 10.2 | |
| OpenScene3DSetting=Privacy-preserving, Input=Depth-only geometry2026.07 | 4.7 | — | — | — | 8.8 | |
| Mosaic3D-DepthOnlySetting=Privacy-preserving, Input=Depth-only geometry2026.07 | 4.5 | — | — | — | 8.7 | |
| D2RGB + OpenSeg FusionSetting=Privacy-preserving, Input=Depth to RGB* + Depth2026.07 | 1.4 | — | — | — | 3.9 |