Research map / Label-efficient and robust learning / Domain adaptation and generalization
Rectifying Pseudo Label Learning via Uncertainty Estimation for Domain Adaptive Semantic Segmentation
Zhedong Zheng, Yi Yang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add variance regularization (exp(-KL) weighted CE + KL) to pseudo label learning | MRNet pseudo label learning · GTA5 -> Cityscapes | mIoU: 48.3 -> 50.3 | Improved |
| add variance regularization to pseudo label learning | MRNet pseudo label learning · SYNTHIA -> Cityscapes (16 classes) | mIoU: 46.5 -> 47.9 | Improved |
| add variance regularization to pseudo label learning | MRNet pseudo label learning · SYNTHIA -> Cityscapes (13 classes) | mIoU*: 53.8 -> 54.9 | Improved |
| add variance regularization to pseudo label learning | MRNet pseudo label learning · Cityscapes -> Oxford RobotCar | mIoU: 73.9 -> 74.4 | Improved |
| replace hand-crafted confidence threshold with variance regularization (ablation) | pseudo label learning · GTA5 -> Cityscapes | mIoU: 48.4 -> 50.3 | Improved |
| raise confidence threshold to >0.99 (ablation) | pseudo label learning · GTA5 -> Cityscapes | mIoU: 48.3 -> 45.5 | Hurt |
| use a moderate threshold >0.90 versus none (ablation) | pseudo label learning · GTA5 -> Cityscapes | mIoU: 48.3 -> 48.4 | Improved |
| use pseudo labels from weaker AdaptSegNet with variance regularization (ablation) | pseudo label learning · GTA5 -> Cityscapes (AdaptSegNet pseudo labels) | mIoU: 46.8 -> 47.4 | Improved |
| add dropout (rate 0.1) in classifiers (ablation) | variance regularization · GTA5 -> Cityscapes | mIoU: 49.6 -> 50.3 | Improved |
| vary dropout rate from 0.1 to 0.7 (ablation) | variance regularization · GTA5 -> Cityscapes | mIoU: 50.3 -> 50.0 | No clear effect |
| use KL(F||F_aux) instead of MSE distance (ablation) | variance regularization · GTA5 -> Cityscapes | mIoU: 49.6 -> 50.3 | Improved |
| use KL(F||F_aux) instead of swapped KL(F_aux||F) (ablation) | variance regularization · GTA5 -> Cityscapes | mIoU: 49.4 -> 50.3 | Improved |
Closest papers by meaning
- Uncertainty-Aware Pseudo Label Refinery for Domain Adaptive Semantic Segmentation (ICCV 2021 · similarity 0.95)
- Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · similarity 0.93)
- Continuous Pseudo-Label Rectified Domain Adaptive Semantic Segmentation With Implicit Neural Representations (CVPR 2023 · similarity 0.92)
- Hard-Aware Instance Adaptive Self-Training for Unsupervised Cross-Domain Semantic Segmentation (TPAMI 2025 · similarity 0.92)
- Generalize Then Adapt: Source-Free Domain Adaptive Semantic Segmentation (ICCV 2021 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
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.
