Research map / Label-efficient and robust learning / Domain adaptation and generalization
Towards Better Stability and Adaptability: Improve Online Self-Training for Model Adaptation in Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| apply DT-ST online self-training to the source-trained model | Semantic segmentation model · GTA5 → Cityscapes | mIoU: 36.8% -> 52.1% | Helped |
| apply DT-ST online self-training to the source-trained model | Semantic segmentation model · SYNTHIA → Cityscapes | mIoU: 38.1% -> 47.7% | Helped |
| apply DT-ST online self-training to the source-trained model | Semantic segmentation model · GTA5 → BDD-100k | average mIoU: 19.7% -> 30.1% | Helped |
| add entropy-feedback dynamic teacher updates (ablation) | Base online self-training · GTA5 → Cityscapes | mIoU: +1.6% | Helped |
| add soft-neighborhood-density feedback dynamic teacher updates (ablation) | Base online self-training · GTA5 → Cityscapes | mIoU: 46.2 -> 47.8 | Helped |
| add training-consistency-based resampling (ablation) | Base online self-training · GTA5 → Cityscapes | mIoU: +4.4% | Helped |
| select resampling candidates using training consistency rather than softmax entropy (ablation) | TCR copy-paste resampling · GTA5 → Cityscapes | mIoU: 52.8 -> 54.4 | Helped |
| combine entropy-feedback dynamic teacher updates with TCR (ablation) | Base online self-training · GTA5 → Cityscapes | mIoU: 50.7 -> 53.8 | Mixed |
| combine soft-neighborhood-density feedback dynamic teacher updates with TCR (ablation) | Base online self-training · GTA5 → Cityscapes | mIoU: 50.6 -> 52.3; 54.4 -> 55.4 | Helped |
| vary the teacher update weight γ from 0.99 to 0.999 (ablation) | DTU · GTA5 → Cityscapes | mIoU | No clear effect |
| vary the meta maximum iteration M (ablation) | DTU · GTA5 → Cityscapes and SYNTHIA → Cityscapes | mIoU: 55.4, 55.3, 55.1, 54.9; 50.7, 50.7, 50.5, 50.1 | No clear effect |
| vary TCR candidate fraction and reliability evaluation interval (ablation) | TCR · GTA5 → Cityscapes | mIoU: 54.1 to 55.5 | No clear effect |
Closest papers by meaning
- Unsupervised Model Adaptation for Continual Semantic Segmentation (AAAI 2021 · similarity 0.93)
- Source-Free Domain Adaptation for Semantic Segmentation (CVPR 2021 · similarity 0.93)
- On the Road to Online Adaptation for Semantic Image Segmentation (CVPR 2022 · similarity 0.92)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.91)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 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.