Research map / Label-efficient and robust learning / Domain adaptation and generalization
Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace offline pseudo labels with online prototypical denoising (ablation) | Self-training with SCE · GTA5→Cityscapes | mIoU: 45.6 to 52.3 | Helped |
| Add SCE to self-training (ablation) | Self-training · GTA5→Cityscapes | mIoU: 45.2 to 45.6; +0.4 | Helped |
| Add target structure learning without prototypical denoising (ablation) | Self-training + SCE · GTA5→Cityscapes | mIoU: 45.6 to 47.6 | Helped |
| Add structure learning alongside prototypical denoising (ablation) | Self-training + SCE + prototypical denoising · GTA5→Cityscapes | mIoU: 52.3 to 53.7 | Helped |
| Add self-distillation to stage 1 (ablation) | ProDA · GTA5→Cityscapes | mIoU: 55.8 to 56.9 | Helped |
| Use self-supervised SimCLRv2 initialization instead of supervised initialization (ablation) | Self-distillation · GTA5→Cityscapes | mIoU: 55.7 to 56.9 | Helped |
| Replace stage-1 initialization with supervised initialization (ablation) | Self-distillation · GTA5→Cityscapes | mIoU: 56.9 to 55.7 | Hurt |
| Apply a third self-distillation stage (ablation) | ProDA with self-supervised initialization · GTA5→Cityscapes | mIoU: 56.9 to 57.5; 0.6 | Helped |
| Add RandAugment and Cutout to crop augmentation (ablation) | Target structure learning · GTA5→Cityscapes | mIoU: 52.7 to 53.7 | Helped |
| Initialize prototypes from source ground-truth labels rather than target pseudo labels (ablation) | Prototypical denoising · GTA5→Cityscapes | mIoU: 53.7 to 53.6 | No clear effect |
| Use prototype-distance temperature 1 rather than more extreme tested temperatures (ablation) | Prototypical denoising · GTA5→Cityscapes | mIoU: 48.8; 52.1; 53.7; 51.9; 47.5; 44.9; 40.9 | Helped |
| Vary the pseudo-label selection threshold (ablation) | Prototypical denoising · GTA5→Cityscapes | mIoU: 53.7; 53.8; 53.7; 53.7; 53.8; 53.7; 53.3 | No clear effect |
Closest papers by meaning
- Class-Balanced Pixel-Level Self-Labeling for Domain Adaptive Semantic Segmentation (CVPR 2022 · similarity 0.92)
- Continuous Pseudo-Label Rectified Domain Adaptive Semantic Segmentation With Implicit Neural Representations (CVPR 2023 · similarity 0.91)
- Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning (CVPR 2023 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- Stable Neighbor Denoising for Source-free Domain Adaptive Segmentation (CVPR 2024 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.