Semantic segmentation on ScanNet20 (val)
77.6mIoUUniPre3D
Evaluation results
| Method | Links | ||
|---|---|---|---|
| UniPre3DBackbone=PTv3, Pre-training=UniPre3D2025.06 | 77.6 | — | |
| PTv3Backbone=PTv3, Pre-training=None2025.06 | 77.5 | — | |
| PonderV2Backbone=SparseUNet, Pre-training=PonderV22025.06 | 77 | — | |
| PPT(Unsup.)Backbone=SparseUNet, Pre-training=PPT(Unsup.)2025.06 | 75.8 | — | |
| UniPre3DBackbone=SparseUNet, Pre-training=UniPre3D2025.06 | 75.8 | — | |
| OctFormerBackbone=OctFormer, Pre-training=None2025.06 | 75.7 | — | |
| MSCBackbone=SparseUNet, Pre-training=MSC2025.06 | 75.5 | — | |
| PTv2Backbone=PTv2, Pre-training=None2025.06 | 75.4 | — | |
| PonderV2+Backbone=SparseUNet, Pre-training=PonderV2† (Single-dataset)2025.06 | 74.6 | — | |
| STBackbone=ST, Pre-training=None2025.06 | 74.3 | — | |
| PCBackbone=SparseUNet, Pre-training=PC2025.06 | 74.1 | — | |
| CSCBackbone=SparseUNet, Pre-training=CSC2025.06 | 73.8 | — | |
| PonderV1Backbone=SparseUNet, Pre-training=PonderV12025.06 | 73.5 | — | |
| SparseUNetBackbone=SparseUNet, Pre-training=None2025.06 | 72.2 | — | |
| PointNextBackbone=PointNext, Pre-training=None2025.06 | 71.5 | — | |
| PTv1Backbone=PTv1, Pre-training=None2025.06 | 70.6 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Evaluation Protocol=Linear probing2026.01 | 67.72 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Resolution=H/4 x W/4, Evaluation Protocol=Linear probing2026.01 | 67.55 | — | |
| EMSANetBackbone=2x ResNet34-NBt1D, Extra Training Data=true2026.01 | 66.96 | — | |
| EMSAFormerBackbone=SwinV2-T-1282026.01 | 64.75 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Resolution=H/4 x W/4, Evaluation Protocol=Visual mean2026.01 | 56.75 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Evaluation Protocol=Visual mean2026.01 | 56.57 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Resolution=H/4 x W/4, Evaluation Protocol=Text-based2026.01 | 50.32 | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Evaluation Protocol=Text-based2026.01 | 50.16 | — |