Semantic Segmentation on NYU v2 (mIoU, SS/FS %)
51.1mIoUFully Supervised
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Fully SupervisedBackbone=Segformer-B4, 3D Supervision=–2026.03 | 51.1 | — | |
| Ours (Recon)Backbone=Segformer-B4, 3D Supervision=Recon.2026.03 | 46.1 | 90.2 | |
| SASFormerBackbone=Segformer-B4, 3D Supervision=–2026.03 | 44.7 | 87.5 | |
| Ours (Real 3D)Backbone=Segformer-B4, 3D Supervision=LiDAR/Depth2026.03 | 44.7 | 87.6 | |
| EMABackbone=Segformer-B4, 3D Supervision=–2026.03 | 42.9 | 84 | |
| TELBackbone=DeepLabV3+, 3D Supervision=–2026.03 | 38.3 | 75 |