Research map / Label-efficient and robust learning / Domain adaptation and generalization
Instance Adaptive Self-Training for Unsupervised Domain Adaptation
Ke Mei, Chuang Zhu, Jiaqi Zou, Shanghang Zhang
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 constant-threshold self-training after warm-up (ablation) | adversarial warm-up model · GTA5 to Cityscapes | mIoU: 43.8 -> 45.1 | Improved |
| add instance adaptive selector (IAS) for pseudo-labels (ablation) | constant-threshold self-training · GTA5 to Cityscapes | mIoU: 45.1 -> 49.8 | Improved |
| add confident-region KLD regularization (ablation) | IAST with IAS · GTA5 to Cityscapes | mIoU: 49.8 -> 50.7 | Improved |
| add ignored-region entropy minimization (ablation) | IAST with IAS and Rc · GTA5 to Cityscapes | mIoU: 50.7 -> 51.5 | Improved |
| add multi-scale testing (ablation) | IAST · GTA5 to Cityscapes | mIoU: 51.5 -> 52.2 | Improved |
| use instance adaptive thresholds instead of class-balanced thresholds (ablation) | IAS (alpha 0.2, beta 0.9) · GTA5 to Cityscapes | mIoU: 47.9 -> 50.5 | Improved |
| add EMA threshold momentum (beta 0 to 0.9) (ablation) | IAS with alpha 0.2 · GTA5 to Cityscapes | mIoU: 49.8 -> 50.5 | Improved |
| raise EMA momentum beta from 0.9 to 0.99 (ablation) | IAS with alpha 0.2 · GTA5 to Cityscapes | mIoU: 50.5 -> 50.0 | Hurt |
| raise pseudo-label proportion alpha from 0.2 to 0.5 (ablation) | IAS with beta 0.9 · GTA5 to Cityscapes | mIoU: 50.5 -> 48.2 | Hurt |
| add hard-class weight decay (gamma 0 to 8) (ablation) | IAS with alpha 0.2 · GTA5 to Cityscapes | mIoU: 50.5 -> 51.5 | Improved |
| increase gamma from 8 to 16 (ablation) | IAST · GTA5 to Cityscapes | mIoU: 51.5 -> 50.9 | Hurt |
| increase lambda_c from 0.05 to 0.1 (with lambda_i 3.0) (ablation) | IAST · GTA5 to Cityscapes | mIoU: 50.6 -> 51.5 | Improved |
Closest papers by meaning
- Hard-Aware Instance Adaptive Self-Training for Unsupervised Cross-Domain Semantic Segmentation (TPAMI 2025 · similarity 0.96)
- Unsupervised Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training (ECCV 2018 · similarity 0.93)
- Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.89)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · similarity 0.88)
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.
