Semantic Segmentation on S3DIS (test)
74.5mIoUSwin3D-L
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Swin3D-LInput=Mesh Sampled Point Cloud, uses_additional_data=true2024.01 | 74.5 | — | — | |
| Masked Scene ModelingBackbone=HUNet, Pre-training=Masked Scene Modeling, Fine-tuning=true2025.04 | 73.2 | — | — | |
| CSCBackbone=SR-UNet, Pre-training=CSC, Fine-tuning=true2025.04 | 72.2 | — | — | |
| MSCBackbone=HUNet, Pre-training=MSC, Fine-tuning=true2025.04 | 72.1 | — | — | |
| Stratified TransformerInput=Mesh Sampled Point Cloud, uses_additional_data=false2024.01 | 72 | — | — | |
| GCBackbone=SR-UNet, Pre-training=GC, Fine-tuning=true2025.04 | 72 | — | — | |
| Point Transformer v2Input=Mesh Sampled Point Cloud, uses_additional_data=false2024.01 | 71.6 | — | — | |
| HUNetBackbone=HUNet, Pre-training=Scratch, Fine-tuning=true2025.04 | 71.3 | — | — | |
| PCBackbone=SR-UNet, Pre-training=PC, Fine-tuning=true2025.04 | 70.3 | — | — | |
| LCPFormerLCP Module=w/ LCP2022.10 | 70.2 | 76.8 | 90.8 | |
| Point Transformer2022.10 | 70 | 76.8 | 90.4 | |
| LCPFormerLCP Module=w/o LCP2022.10 | 69.3 | 75.2 | 90.2 | |
| DenseKPNET2022.10 | 68.9 | — | 90.8 | |
| ODIN-Swin-BInput=RGBD Point Cloud, uses_additional_data=true2024.01 | 68.6 | — | — | |
| SR-UNetBackbone=SR-UNet, Pre-training=Scratch, Fine-tuning=true2025.04 | 68.2 | — | — | |
| PatchFormer2022.10 | 68.1 | — | — | |
| JSENet2020.07 | 67.7 | — | — | |
| DeepViewAggInput=RGBD Point Cloud, uses_additional_data=false2024.01 | 67.2 | — | — | |
| KPConvType=deform2020.07 | 67.1 | — | — | |
| KPConv2022.10 | 67.1 | 72.8 | — | |
| ODIN-ResNet50Input=RGBD Point Cloud, uses_additional_data=true2024.01 | 66.8 | — | — | |
| KPConvType=rigid2020.07 | 65.4 | — | — | |
| MinkowskiNet2020.07 | 65.4 | — | — | |
| VMVFInput=RGBD Point Cloud, uses_additional_data=false2024.01 | 65.4 | — | — | |
| LocalTransformer2022.10 | 64.1 | 71.9 | 87.6 | |
| PSNet2022.10 | 62.9 | — | 87.8 | |
| MVPNet2020.07 | 62.4 | — | — | |
| RandLA-Net2022.10 | 62.4 | 71.4 | 87.2 | |
| MVPNetInput=RGBD Point Cloud, uses_additional_data=false2024.01 | 62.4 | — | — | |
| HPEIN2020.07 | 61.9 | — | — | |
| ODIN-ResNet50Input=RGBD Point Cloud, uses_additional_data=false2024.01 | 59.7 | — | — | |
| SPH3D-GCN2020.07 | 59.5 | — | — | |
| Param Conv2020.07 | 58.3 | — | — | |
| SPGraph2020.07 | 58 | — | — | |
| SPGraph2022.10 | 58 | 66.5 | 86.4 | |
| PointCNN2020.07 | 57.3 | — | — | |
| RNN Fusion2020.07 | 53.4 | — | — | |
| Tangent ConvolutionsInput signals=Depth (D) + Height (H) + Normals (N) + Color (RGB)2018.07 | 52.8 | 62.2 | 82.5 | |
| TrajConv2022.10 | 52.8 | 62.2 | 82.5 | |
| TangentConv2020.07 | 52.6 | — | — | |
| Tangent ConvolutionsInput signals=Depth (D) + Height (H) + Normals (N)2018.07 | 51.7 | 61 | 82.2 | |
| Tangent ConvolutionsInput signals=Depth (D) + Height (H)2018.07 | 50 | 60 | 81.2 | |
| Tangent ConvolutionsInput signals=Depth (D)2018.07 | 49.8 | 60.3 | 80.2 | |
| PointNet2018.07 | 41.3 | 49.5 | 78.8 | |
| PointNet2022.10 | 41.1 | 23.7 | — | |
| OctNet2018.07 | 26.3 | 39 | 68.9 | |
| ScanNet2018.07 | 24.6 | 35 | 64.2 |