Research map / Label-efficient and robust learning / Domain adaptation and generalization
CDAC: Cross-domain Attention Consistency in Transformer for Domain Adaptive Semantic Segmentation
Kaihong Wang, Donghyun Kim, Rogerio Feris, Margrit Betke
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 cross-domain prediction consistency loss (T2S and S2T) (ablation) | DAFormer · GTAV-to-Cityscapes | mIoU: 68.3 -> 69.1 | Improved |
| add cross-domain attention consistency loss alone (ablation) | DAFormer · GTAV-to-Cityscapes | mIoU: 68.3 -> 68.6 | Improved |
| add attention consistency loss on top of prediction consistency (ablation) | DAFormer + CDAC · GTAV-to-Cityscapes | mIoU: 69.1 -> 69.6 | Improved |
| add attention consistency loss reduces seed variance (ablation) | DAFormer · GTAV-to-Cityscapes | mIoU std: 0.7 -> 0.2 | Improved |
| stop gradient on query vectors in cross-domain branches (ablation) | CDAC · GTAV-to-Cityscapes | mIoU: 69.1 -> 69.6 | Improved |
| perturb self-attention with random Gaussian noise (ablation) | DAFormer · GTAV-to-Cityscapes | mIoU: 68.3 -> 67.3 | Hurt |
| inject uniform attention (context broadcasting) (ablation) | DAFormer · GTAV-to-Cityscapes | mIoU: 68.3 -> 66.9 | Hurt |
| inject uniform attention (context broadcasting) (ablation) | DAFormer · Synthia-to-Cityscapes | mIoU: 60.9 -> 59.1 | Hurt |
| add CDAC to HRDA | HRDA · Synthia-to-Cityscapes | mIoU: 65.8 -> 68.7 | Improved |
| add CDAC to DAFormer | DAFormer · Cityscapes-to-ACDC (test) | mIoU: 55.4 -> 56.5 | Improved |
| add CDAC to Mean-Teacher baseline | Mean-Teacher · GTAV-to-Cityscapes | accuracy: 39.0% -> 41.18% | Improved |
| add CDAC to DACS | DACS · GTAV-to-Cityscapes | accuracy: 52.54% -> 53.79% | 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 3 of 3 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
- Self-Ensembling Attention Networks: Addressing Domain Shift for Semantic Segmentation (AAAI 2019 · similarity 0.89)
- Cross-Modal Learning for Domain Adaptation in 3D Semantic Segmentation (TPAMI 2022 · similarity 0.89)
- Cross-Domain Grouping and Alignment for Domain Adaptive Semantic Segmentation (AAAI 2021 · similarity 0.89)
- Discover, Hallucinate, and Adapt: Open Compound Domain Adaptation for Semantic Segmentation (NeurIPS 2020 · similarity 0.89)
- Unified Domain Adaptive Semantic Segmentation (TPAMI 2025 · similarity 0.88)
- Addressing Domain Gap via Content Invariant Representation for Semantic Segmentation (AAAI 2021 · 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.
