Research map / Label-efficient and robust learning / Domain adaptation and generalization
Exploiting Diverse Characteristics and Adversarial Ambivalence for Domain Adaptive Segmentation
Bowen Cai, Huan Fu, Rongfei Jia, Binqiang Zhao, Hua Li, Yinghui Xu
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 |
|---|---|---|---|
| Apply DCAA instead of the strongest listed baseline | GTA5 → Cityscapes* adaptation · Cityscapes* | mIoU: 45.4 -> 51.4 | Improved |
| Apply DCAA instead of the strongest listed baseline | SYNTHIA → Cityscapes* adaptation · Cityscapes* | mIoU: 42.5 -> 47.4 | Improved |
| Use DCAA rather than OCDA++ | GTA5 → BDD100K · BDD100K Avg. | mIoU: 25.7 -> 34.7 | Improved |
| Replace Mix-Head, Sep-Head, or SM-Head with CAM (ablation) | DCAA · GTA5 → Cityscapes* | mIoU (Stage 1): 46.0 -> 47.2; 45.3 -> 47.2; 45.6 -> 47.2 | Improved |
| Replace CAM with Sep-Head or SM-Head (ablation) | DCAA · GTA5 → Cityscapes* | mIoU (Stage 1): 47.2 -> 45.3; 47.2 -> 45.6 | Hurt |
| Use progressive condition-specific adversarial training rather than standard adversarial training (ablation) | DCAA · GTA5 → Cityscapes* | mIoU: 47.8 -> 51.4 | Improved |
| Use attentive pseudo-label assignment rather than simple voting (ablation) | DCAA-S1 · GTA5 → Cityscapes* | mIoU: 48.7 -> 49.8 | Improved |
| Weight target pseudo-label cross-entropy with discriminator confidence (ablation) | DCAA self-training with λp = 0.6 · GTA5 → Cityscapes* | mIoU: 48.3 -> 50.6 | Improved |
| Add adversarial loss for hard adaptation regions (ablation) | DCAA self-training with λp = 0.6 · GTA5 → Cityscapes* | mIoU: 50.6 -> 51.4 | Improved |
| Use condition-guided style transfer instead of CycleGAN (ablation) | Mix-Head adaptation model · GTA5 → Cityscapes* | mIoU (Stage 1): 42.7 -> 46.0 | Improved |
| Use MUNIT instead of CycleGAN for translation (ablation) | Mix-Head adaptation model · GTA5 → Cityscapes* | mIoU (Stage 1): 42.7 -> 43.4 | Improved |
| Replace CGST with MUNIT or StarGAN translation (ablation) | DCAA Stage 2 · GTA5 → Cityscapes* | mIoU: 45.6 -> 51.4; 48.3 -> 51.4 | Hurt |
Closest papers by meaning
- Category-Level Adversarial Adaptation for Semantic Segmentation using Purified Features (TPAMI 2021 · similarity 0.91)
- An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation (AAAI 2020 · similarity 0.91)
- Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · similarity 0.90)
- Multi-Target Adversarial Frameworks for Domain Adaptation in Semantic Segmentation (ICCV 2021 · similarity 0.90)
- Multi-source Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · similarity 0.90)
- Exploring Weather-aware Aggregation and Adaptation for Semantic Segmentation under Adverse Conditions (ICCV 2025 · 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.
