Research map / Label-efficient and robust learning / Domain adaptation and generalization
Efficient Active Domain Adaptation for Semantic Segmentation by Selecting Information-rich Superpixels
Yuan Gao, Zilei Wang, Yixin Zhang, Bohai Tu
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 |
|---|---|---|---|
| replace ASAL fusion with low-uncertainty superpixel fusion (ablation) | superpixel active domain adaptation · GTA5->Cityscapes | mIoU upper bound: 69.11% -> 72.58% | Improved |
| fuse low-uncertainty superpixels (873 clicks) versus labeling all original superpixels (4,985) (ablation) | DeepLabv2 · GTA5->Cityscapes | mIoU: 73.1 -> 72.6 | No clear effect |
| use two-stage acquisition (A1 large merged superpixels then A2) instead of single-stage DS (ablation) | DeepLabv2 at 40 clicks · GTA5->Cityscapes | mIoU: 58.5 -> 68.7 | Improved |
| use two-stage acquisition instead of ASAL (ablation) | DeepLabv2 at 40 clicks · GTA5->Cityscapes | mIoU: 57.0 -> 68.7 | Improved |
| use domain-difference selection in phase 2 versus other selection algorithms (ablation) | A2 · GTA5->Cityscapes | mIoU: +0.41% to +0.65% over second place | Improved |
| add cross-domain mixing (ablation) | baseline supervised training · GTA5->Cityscapes (40 clicks) | mIoU: 63.9 -> 67.4 | Improved |
| add pseudo labels on target domain (ablation) | M(1) · GTA5->Cityscapes (40 clicks) | mIoU: 67.4 -> 68.0 | Improved |
| add consistency regularization on domain-mix samples (ablation) | M(2) · GTA5->Cityscapes (40 clicks) | mIoU: 68.0 -> 68.7 | Improved |
| apply training techniques to RIPU (ablation) | RIPU (11,535 clicks) · GTA5->Cityscapes | mIoU: 62.8 -> 71.9 | Improved |
| use superpixel-level labeling with 640 clicks versus pixel-level 26,215 clicks | DeepLabv3+ · GTA5->Cityscapes | mIoU: 71.3 -> 72.9 | Improved |
Closest papers by meaning
- Revisiting Superpixels for Active Learning in Semantic Segmentation With Realistic Annotation Costs (CVPR 2021 · similarity 0.90)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.90)
- Source-Free Domain Adaptation for Semantic Segmentation (CVPR 2021 · similarity 0.90)
- I2F: A Unified Image-to-Feature Approach for Domain Adaptive Semantic Segmentation (TPAMI 2022 · similarity 0.89)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.89)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · similarity 0.88)
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.
