Instance Segmentation on Cityscapes 10% labeled data (val)
33.9maskAPCAST (student refinement)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CAST (student refinement)Data Regime=Labeled only, Model Stage=Student Distillation, Parameters=52M2025.05 | 33.9 | 58.7 | |
| CAST (knowledge transfer)Data Regime=Labeled+Unlabeled, Model Stage=Student Distillation, Parameters=52M2025.05 | 32.2 | 56.5 | |
| Guided dist.Data Regime=Labeled+Unlabeled, Model Stage=Student Distillation, Parameters=52M2025.05 | 30.8 | 52.9 | |
| CAST (teacher adaptation)Data Regime=Labeled+Unlabeled, Model Stage=Teacher Adaptation, Parameters=568M2025.05 | 30.5 | 56.6 | |
| Unbiased Teacher*Data Regime=Labeled+Unlabeled, Model Stage=Teacher Adaptation, Parameters=568M2025.05 | 29.8 | 54.9 | |
| Self-training*Data Regime=Labeled+Unlabeled, Model Stage=Teacher Adaptation, Parameters=568M2025.05 | 29.7 | 54.9 | |
| Supervised fine-tuningData Regime=Labeled only, Model Stage=Teacher Adaptation, Parameters=568M2025.05 | 28.7 | 53.4 | |
| Vemulapalli et al.*Data Regime=Unlabeled only, Model Stage=Student Distillation, Parameters=52M2025.05 | 24.4 | 45.6 | |
| PAISData Regime=Labeled+Unlabeled, Model Stage=Student Distillation, Parameters=52M2025.05 | 22.9 | 44.9 | |
| Zero-shot VFMData Regime=None (pretrained), Model Stage=Teacher Adaptation, Parameters=568M2025.05 | 22 | 42.3 | |
| Supervised fine-tuningData Regime=Labeled only, Model Stage=Student Distillation, Parameters=52M2025.05 | 21.1 | 38.7 |