Semantic Segmentation on GWFSS 1.0 (test)
93.11mIoUDepthCropSeg++
Evaluation Results
| Method | Links | |
|---|---|---|
| DepthCropSeg++Depth-based Pseudo-masks=true, Two-stage Self-training=true, FADE Upsampling=true2026.01 | 93.11 | |
| DepthCropSeg++Depth-based Pseudo-masks=true, Two-stage Self-training=true, FADE Upsampling=false2026.01 | 92.88 | |
| DepthCropSeg++Depth-based Pseudo-masks=true, Two-stage Self-training=false, FADE Upsampling=false2026.01 | 92.86 | |
| Fully Supervised2026.01 | 92.75 | |
| Semi-supervised baselineDepth-based Pseudo-masks=false, Two-stage Self-training=false, FADE Upsampling=false2026.01 | 92.36 | |
| Proposed Self-Training FrameworkBackbone=SegFormer, Training Strategy=Two-Stage Hybrid (Iterative Teacher-Student Loop), Cross-validation=10-fold, Model Ensembling=true, Test-Time Augmentation (TTA)=true2025.12 | 70.99 | |
| GWFSS Top-Solution2026.01 | 69.5 | |
| HQ-SAM2026.01 | 44.54 | |
| SAM2026.01 | 32.8 |