Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning Pseudo-Relations for Cross-domain Semantic Segmentation
Dong Zhao, Shuang Wang, Qi Zang, Dou Quan, Xiutiao Ye, Rui Yang, Licheng Jiao
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 pseudo-relation learning (L_pr) on top of SAC + copy-paste (ablation) | SAC self-training · GTA5 -> Cityscapes | mIoU: 55.8 -> 59.6 | Improved |
| use threshold-free L_pr instead of thresholded L_pr-hat (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 58.9 -> 59.6 | Improved |
| also perform relation learning on source domain (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 59.6 -> 59.6 | No clear effect |
| add distillation retraining after RTea (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 59.6 -> 61.9 | Improved |
| fuse low-level RGBXY relations with high-level pseudo-label relations (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 56.8 / 57.8 -> 59.6 | Improved |
| include reliable-unreliable pixel pairs in relation learning (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 57.2 -> 59.3 (adding R-U to R-R) | Improved |
| add unreliable-unreliable pairs (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 59.3 -> 59.6 | Improved |
| restrict relations to local grids instead of global image (ablation) | RTea · GTA5 -> Cityscapes | mIoU: global R-U 51.4 (-4.4) vs local grid R-U 57.9 (+2.1) | Improved |
| apply RTea on different pseudo-label selection strategies (PD, CD, AD, HT) (ablation) | online self-training baselines · GTA5 -> Cityscapes | mIoU: 55.0 -> 58.8 (PD); 54.4 -> 56.8 (CD); 53.2 -> 54.8 (AD); 55.8 -> 59.6 (HT) | Improved |
| vary threshold theta and decay beta (ablation) | RTea · GTA5 -> Cityscapes | mIoU: 58.8 to 59.6 across settings | No clear effect |
| use the same Gaussian metric for low- and high-level relations (ablation) | RTea · GTA5 -> Cityscapes | mIoU / time per batch: 59.6 -> 59.6 mIoU; 40ms -> 29ms | No clear effect |
| add RTea to Daformer and HRDA backbones | Daformer, HRDA · GTA5 -> Cityscapes | mIoU: 68.3 -> 70.6 (Daformer); 73.8 -> 74.7 (HRDA) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Daformer — DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (CVPR 2022) · better on 2 of 2 shared results
- HRDA — HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation (ECCV 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contextual-Relation Consistent Domain Adaptation for Semantic Segmentation (ECCV 2020 · similarity 0.89)
- Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · similarity 0.88)
- Feature Re-Representation and Reliable Pseudo Label Retraining for Cross-Domain Semantic Segmentation (TPAMI 2022 · similarity 0.88)
- Pay Attention to Target: Relation-Aware Temporal Consistency for Domain Adaptive Video Semantic Segmentation (AAAI 2024 · similarity 0.88)
- Connectivity-Driven Pseudo-Labeling Makes Stronger Cross-Domain Segmenters (NeurIPS 2024 · similarity 0.88)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.87)
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.
