Research map / Adversarial robustness and security / Attacks on text and graphs
Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation
Xiaogang Xu, Hengshuang Zhao, 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 |
|---|---|---|---|
| add standard adversarial training (SAT) | PSPNet · VOC, white-box BIM L_inf attack | mIoU BIM n=6: 5.4 -> 27.9 | Improved |
| add standard adversarial training (SAT) | DeepLabv3 · VOC, white-box BIM L_inf attack | mIoU BIM n=6: 5.5 -> 28.8 | Improved |
| add standard adversarial training (SAT) | PSPNet · VOC, clean samples | mIoU: 76.9 -> 74.3 | Hurt |
| replace SAT with DDC-AT | PSPNet · VOC, clean samples | mIoU: 74.3 -> 76.0 | Improved |
| replace SAT with DDC-AT | DeepLabv3 · VOC, clean samples | mIoU: 72.7 -> 75.2 | Improved |
| replace SAT with DDC-AT | PSPNet · VOC, white-box BIM L_inf attack | mIoU BIM n=6: 27.9 -> 33.8 | Improved |
| replace SAT with DDC-AT | DeepLabv3 · VOC, white-box BIM L_inf attack | mIoU BIM n=6: 28.8 -> 32.3 | Improved |
| replace SAT with DDC-AT | PSPNet · Cityscapes, white-box BIM L_inf attack | mIoU BIM n=6: 25.8 -> 28.7 | Improved |
| replace SAT with DDC-AT | DeepLabv3 · Cityscapes, white-box BIM L_inf attack | mIoU BIM n=6: 26.2 -> 29.0 | Improved |
| replace SAT with DDC-AT | PSPNet · Cityscapes, clean samples | mIoU: 69.0 -> 71.7 | Improved |
| use DDC-AT division with clean boundary pixels in auxiliary branch (DDC-AT-M) instead of the full dynamic division (ablation) | DDC-AT · VOC ablation, PSPNet, BIM n=6 | mIoU: 33.8 -> 30.4 | Hurt |
| use DDC-AT-N division instead of the full DDC-AT division (ablation) | DDC-AT · VOC ablation, PSPNet, BIM n=6 | mIoU: 33.8 -> 31.8 | Hurt |
Closest papers by meaning
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks (CVPR 2018 · similarity 0.92)
- Towards Reliable Evaluation and Fast Training of Robust Semantic Segmentation Models (ECCV 2024 · similarity 0.91)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.91)
- Proximal Splitting Adversarial Attack for Semantic Segmentation (CVPR 2023 · similarity 0.90)
- Rethinking Adversary in Semantic Segmentation: An Out-of-Distribution Perspective (ECCV 2026 · similarity 0.89)
- RP-PGD: Boosting Segmentation Robustness with a Region-and-Prototype Based Adversarial Attack (AAAI 2025 · similarity 0.87)
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.
