Research map / Label-efficient and robust learning / Domain adaptation and generalization
Focus on Your Target: A Dual Teacher-Student Framework for Domain-Adaptive Semantic Segmentation
Xinyue Huo, Lingxi Xie, Wengang Zhou, Houqiang Li, Qi Tian
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 |
|---|---|---|---|
| increase the proportion of target-domain data in a single teacher-student model (<T,T> mixing) (ablation) | DAFormer · GTAv->Cityscapes | mIoU: 68.3 -> 67.2 | Hurt |
| use the dual teacher-student framework together with more target data (ablation) | DAFormer · GTAv->Cityscapes | mIoU: 67.2 -> 69.7 | Improved |
| add bidirectional learning on top of DTS with Focus on Your Target (ablation) | DTS (DAFormer) · GTAv->Cityscapes | mIoU: 69.7 -> 71.2 | Improved |
| add Focus on Your Target (more target data) on top of DTS with bidirectional learning (ablation) | DTS (DAFormer) · SYNTHIA->Cityscapes (16 classes) | mIoU: 61.2 -> 63.7 | Improved |
| add DTS to MIC, HRDA and DAFormer baselines | MIC/HRDA/DAFormer · GTAv->Cityscapes | mIoU: +2.9, +1.4, +0.6 | Improved |
| choose Setting A vs B using pseudo-label confidence criterion (ablation) | DTS (DAFormer) · GTAv->Cityscapes vs SYNTHIA->Cityscapes | mIoU: GTAv B 71.2 vs A 70.3; SYNTHIA A 63.8 vs B 61.2 | Mixed |
| each group uses only the other group's teacher for pseudo labels (ablation) | DTS (DAFormer) · GTAv->Cityscapes | mIoU: 68.3 -> 68.5 | No clear effect |
| use student f1,t+1 st instead of teacher f1,t+1 te for group 2 pseudo labels (learning from future) (ablation) | DTS (DAFormer) · GTAv->Cityscapes | mIoU: 69.5 -> 71.3 | Improved |
| apply DTS to CNN backbone (ResNet101 DeepLabV2) | DACS and DAFormer · GTAv->Cityscapes (ResNet101) | mIoU: 53.9 -> 55.2; 56.0 -> 59.9 | Improved |
| apply DTS to semi-supervised segmentation vs Mean Teacher | PASCAL VOC DeepLabV3+ · PASCAL VOC semi-supervised | mIoU: +3.5, +1.7, +0.8 (1/16, 1/8, 1/4) | Improved |
| use only <T,T> or only <S,T> mixing for the second group (ablation) | DTS (DAFormer) · SYNTHIA->Cityscapes (16 classes) | mIoU: 63.8 -> 61.1 / 59.4 | Hurt |
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
- Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation (ICCV 2021 · similarity 0.93)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.93)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.93)
- Unsupervised Model Adaptation for Continual Semantic Segmentation (AAAI 2021 · similarity 0.93)
- Unsupervised Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training (ECCV 2018 · similarity 0.92)
- DAST: Unsupervised Domain Adaptation in Semantic Segmentation Based on Discriminator Attention and Self-Training (AAAI 2021 · similarity 0.92)
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.
