3D Semantic Segmentation on S3DIS
72.2mIoUHou et al.
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Hou et al.Pre-trained=Yes2023.05 | 72.2 | — | |
| PointContrastPre-trained=Yes2023.05 | 70.9 | 77 | |
| DepthContrastPre-trained=Yes2023.05 | 70.6 | — | |
| Bridge3DPre-trained=Yes2023.05 | 70.2 | 76.1 | |
| PViT+Pix4PointPre-trained=Yes2023.05 | 69.6 | 75.2 | |
| SR-UNetPre-trained=None2023.05 | 68.2 | 75.5 | |
| PLOVIS2026.04 | 66 | 77.6 | |
| PointCNN2018.01 | 65.39 | — | |
| Point-MAEPre-trained=Yes2023.05 | 64.8 | 70.2 | |
| PVITPre-trained=None2023.05 | 64.4 | 69.9 | |
| ERDA2026.04 | 64 | 74.2 | |
| AAD-Net2026.04 | 63.4 | 72.8 | |
| DG-Net2026.04 | 63.4 | 72.8 | |
| MLP probingFine-tuning strategy=MLP probing2026.04 | 63.3 | 73.2 | |
| DGCNN (Ours)Pre-training=Self-supervised (Ours), Few-shot Setting=5-way 10-shot, Backbone=DGCNN2020.09 | 62.63 | 68.87 | |
| DGCNN (VoxelSSL)Pre-training=VoxelSSL, Few-shot Setting=5-way 10-shot, Backbone=DGCNN2020.09 | 62.42 | 64.59 | |
| Decoder probingFine-tuning strategy=Decoder probing2026.04 | 61.6 | 72.2 | |
| DGCNN (Random Init)Pre-training=None, Few-shot Setting=5-way 10-shot, Backbone=DGCNN2020.09 | 61.36 | 67.41 | |
| Bridge3DPre-trained=None2023.05 | 61.1 | 67.2 | |
| PointBertPre-trained=Yes2023.05 | 60.8 | 69.9 | |
| Full fine-tuningFine-tuning strategy=Full fine-tuning2026.04 | 60.5 | 71.3 | |
| Standard TransformerPre-trained=None2023.05 | 60 | 68.6 | |
| DGCNN2018.01 | 56.1 | 84.1 | |
| G + RCUfeatures=grid-blocks with recurrent consolidation units2018.01 | 49.7 | 81.1 | |
| MS + CU(2)features=multi-scale block features with consolidation units2018.01 | 47.8 | 79.2 | |
| PointNet2018.01 | 47.6 | 78.5 | |
| PointNettype=baseline2018.01 | 20.1 | 53.2 |