Research map / Label-efficient and robust learning / Domain adaptation and generalization
DecoupleNet: Decoupled Network for Domain Adaptive Semantic Segmentation
Xin Lai, Zhuotao Tian, Xiaogang Xu, Yingcong Chen, Shu Liu, Hengshuang Zhao, Liwei Wang, Jiaya Jia
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 |
|---|---|---|---|
| use decoupled network architecture with shallow-feature alignment instead of AdaptSegNet (ablation) | AdaptSegNet · GTA5->Cityscapes | mIoU: +3.0% mIoU | Improved |
| add Self-Discrimination (SD) (ablation) | DecoupleNet · GTA5->Cityscapes | mIoU: 47.1 -> 49.0 | Improved |
| add Self-Discrimination (SD) (ablation) | AdaptSegNet · GTA5->Cityscapes | mIoU: 44.1 -> 46.0 | Improved |
| add shallow-feature adversarial loss L_adv_low (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: +2.2% mIoU | Improved |
| add output-space adversarial loss L_adv (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: 48.7 -> 49.0 | Improved |
| add auxiliary classifier for SD (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: 46.0 -> 49.0 | Improved |
| add class-wise thresholds for pseudo labels (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: 48.4 -> 49.0 | Improved |
| align source to target (phi_s -> phi_t) rather than target to source (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: 47.3 -> 49.0 | Improved |
| add shallow-feature alignment versus none (ablation) | DecoupleNet + SD · GTA5->Cityscapes | mIoU: 46.8 -> 49.0 | Improved |
| decouple layer0 and layer1 instead of layer0 only or layer0-2 (ablation) | DecoupleNet · GTA5->Cityscapes | mIoU: 47.7 / 49.0 / 47.9 | Mixed |
| use averaged full-image predictions versus fixed pseudo soft labels in OEST (ablation) | OEST · GTA5->Cityscapes | mIoU: 55.6 -> 56.7 | Improved |
| average with full-image predictions instead of crop predictions in OEST (ablation) | OEST · GTA5->Cityscapes | mIoU: 55.2 -> 56.7 | Improved |
Closest papers by meaning
- DAST: Unsupervised Domain Adaptation in Semantic Segmentation Based on Discriminator Attention and Self-Training (AAAI 2021 · similarity 0.92)
- Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · similarity 0.91)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.90)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.89)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · 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.
