Research map / Label-efficient and robust learning / Domain adaptation and generalization
Unsupervised Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training
Yang Zou, Zhiding Yu, B.V.K. Vijaya Kumar, Jinsong Wang
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 class-balanced pseudo-label selection (CBST) to self-training (ablation) | ST, ResNet-38 · GTA5 to Cityscapes | mIoU (%): 41.5 -> 45.2 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, FCN8s-VGG16 · GTA5 to Cityscapes | mIoU (%): 28.1 -> 30.9 | Improved |
| add spatial priors to CBST (ablation) | CBST, FCN8s-VGG16 · GTA5 to Cityscapes | mIoU (%): 30.9 -> 36.1 | Improved |
| add spatial priors to CBST (ablation) | CBST, ResNet-38 · GTA5 to Cityscapes | mIoU (%): 45.2 -> 46.2 | Improved |
| add multi-scale testing (0.5,0.75,1.0) (ablation) | CBST-SP, ResNet-38 · GTA5 to Cityscapes | mIoU (%): 46.2 -> 47.0 | Improved |
| add self-training (ST) over source-only model | ResNet-38 · GTA5 to Cityscapes | mIoU (%): 35.4 -> 41.5 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, ResNet-38 · SYNTHIA to Cityscapes (16 classes) | mIoU (%): 32.2 -> 42.5 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, FCN8s-VGG16 · SYNTHIA to Cityscapes (16 classes) | mIoU (%): 23.9 -> 35.4 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, ResNet-38 · Cityscapes to NTHU (Rome) | mIoU (%): 48.9 -> 53.6 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, ResNet-38 · Cityscapes to NTHU (Taipei) | mIoU (%): 44.6 -> 50.3 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, ResNet-38 · Cityscapes to NTHU (Tokyo) | mIoU (%): 44.6 -> 48.8 | Improved |
| add class-balanced selection (CBST) to self-training (ablation) | ST, ResNet-38 · Cityscapes to NTHU (Rio) | mIoU (%): 50.8 -> 52.2 | Improved |
Closest papers by meaning
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.95)
- DAST: Unsupervised Domain Adaptation in Semantic Segmentation Based on Discriminator Attention and Self-Training (AAAI 2021 · similarity 0.93)
- Instance Adaptive Self-Training for Unsupervised Domain Adaptation (ECCV 2020 · similarity 0.93)
- Hard-Aware Instance Adaptive Self-Training for Unsupervised Cross-Domain Semantic Segmentation (TPAMI 2025 · similarity 0.93)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.92)
- Towards Unsupervised Domain Bridging via Image Degradation in Semantic Segmentation (NeurIPS 2025 · similarity 0.91)
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.
