Research map / Label-efficient and robust learning / Domain adaptation and generalization
Prototypical Contrast Adaptation for Domain Adaptive Semantic Segmentation
Zhengkai Jiang, Yuxi Li, Ceyuan Yang, Peng Gao, Yabiao Wang, Ying Tai, Chengjie 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 prototypical contrast adaptation (feature+output) (ablation) | source-only DeepLab-v2 · GTA5->Cityscapes | mIoU: 37.3 -> 48.8 | Improved |
| add ProCA before self-training (ablation) | adaptive self-training · GTA5->Cityscapes | mIoU: 43.9 -> 55.1 | Improved |
| add class-wise adaptive threshold self-training (ablation) | ProCA · GTA5->Cityscapes | mIoU: 48.8 -> 55.1 | Improved |
| add multi-scale testing (ablation) | ProCA + adaptive ST · GTA5->Cityscapes | mIoU: 55.1 -> 56.3 | Improved |
| add output-level to feature-level adaptation (multi-level) (ablation) | ProCA · GTA5->Cityscapes | mIoU: 47.9 -> 48.8 | Improved |
| use ProCA instead of FADA (ablation) | source-only · GTA5->Cityscapes | mIoU: 46.9 -> 48.8 | Improved |
| use ProCA instead of pixel memory bank (ablation) | source-only · GTA5->Cityscapes | mIoU: 47.6 -> 48.8 | Improved |
| use mixed source-target prototype updating instead of fixed prototypes (ablation) | ProCA · GTA5->Cityscapes | mIoU: 47.8 -> 48.8 | Improved |
| add cross-domain t->s contrast to source-source contrast (ablation) | ProCA · GTA5->Cityscapes | mIoU: 44.9 -> 48.8 | Improved |
| use adaptive class-wise thresholds instead of naive 0.9 threshold (ablation) | ProCA self-training · GTA5->Cityscapes | mIoU: 55.2 -> 56.3 | Improved |
| use ProDA prototype-based pseudo labels instead of adaptive thresholds (ablation) | ProCA self-training · GTA5->Cityscapes | mIoU: 56.3 -> 57.5 | Improved |
| raise pseudo-label percentage eta from 60 to 90 (ablation) | ProCA self-training · GTA5->Cityscapes | mIoU: 55.1 -> 52.5 | Hurt |
Closest papers by meaning
- Bi-directional Contrastive Learning for Domain Adaptive Semantic Segmentation (ECCV 2022 · similarity 0.93)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.91)
- Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning (CVPR 2023 · similarity 0.91)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.91)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation (AAAI 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.
