3D Semantic Segmentation on ScanNet20 v2 (test)
79.8mIoUPTv3 + PPT (Ours)
Evaluation results
| Method | Links | |
|---|---|---|
| PTv3 + PPT (Ours)Backbone=PTv3, Training Data=ScanNet + ScanNet200 + ScanNet++ + Structure3D + ARKit LabelMaker2024.10 | 79.8 | |
| PTv3 + PPTBackbone=PTv3, Training Data=ScanNet + S3DIS + Structure3D2024.10 | 79.4 | |
| PTv3 + fine-tune (Ours)Backbone=PTv3, Training Data=ARKit LabelMaker → ScanNet2024.10 | 79 | |
| MinkUNet + Mix3DBackbone=MinkowskiNet, Training Data=ScanNet2024.10 | 78.1 | |
| PTv3 (vanilla)Backbone=PTv3, Training Data=ScanNet2024.10 | 77.9 | |
| LSK3DNetInput=PV2024.03 | 75.5 | |
| Point Transformer v2Input=P2024.03 | 75.2 | |
| LargeKernel3DInput=V2024.03 | 73.9 | |
| Stratified TransformerInput=P2024.03 | 73.7 | |
| MinkUNetInput=V2024.03 | 73.6 | |
| MinkUNet (vanilla)Backbone=MinkowskiNet, Training Data=ScanNet2024.10 | 73.6 | |
| SparseConvNetInput=V2024.03 | 72.5 | |
| JSENetInput=P2024.03 | 69.9 | |
| FusionNetInput=P2024.03 | 68.8 | |
| KPConvInput=P2024.03 | 68.6 | |
| PointConvInput=P2024.03 | 66.6 | |
| PointASNLInput=P2024.03 | 66.6 | |
| RandLA-NetInput=P2024.03 | 64.5 | |
| JointPointBasedInput=P2024.03 | 63.4 | |
| SegGCNInput=P2024.03 | 58.9 | |
| PointNet++Input=P2024.03 | 55.7 | |
| PanopticFusionInput=P2024.03 | 52.9 | |
| 3DMVInput=P2024.03 | 48.4 | |
| PointCNNInput=P2024.03 | 45.8 |