Research map / Adversarial robustness and security / Attacks on text and graphs
Rethinking Adversary in Semantic Segmentation: An Out-of-Distribution Perspective
Yuxuan Zhang, Shuchang Wang, Xiaoman Liu, Zhenbo Shi, Zhidong Yu, Wei Song, Wei Yang
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 FWS and WKS objectives to prediction-based attack (ablation) | SegPGD attack · Pascal VOC, SegPGD3-AT PSPNet | mIoU under attack (lower = stronger): 26.48 -> 19.82 | Improved |
| add FWS and WKS objectives to prediction-based attack (ablation) | SegPGD attack · Pascal VOC, CosPGD3-AT DeepLabv3 | mIoU under attack (lower = stronger): 25.88 -> 16.57 | Improved |
| use F-SLIC instead of category prototypes (ablation) | SegOOD · Pascal VOC, SegPGD3-AT PSPNet | mIoU under attack (lower = stronger): 34.85 -> 19.82 | Improved |
| use F-SLIC instead of RGB SLIC (ablation) | SegOOD · Pascal VOC, SegPGD3-AT PSPNet | mIoU under attack (lower = stronger): 24.91 -> 19.82 | Improved |
| add SegOOD adversarial samples (separate) in AT | SegPGD3-AT · Pascal VOC PSPNet, SegOOD attack | mIoU: 19.82 -> 30.15 | Improved |
| add SegOOD adversarial samples (separate) in AT | SegPGD3-AT · Pascal VOC PSPNet, SegPGD attack | mIoU: 27.09 -> 30.37 | Improved |
| add SegOOD samples in AT | SegPGD3-AT · Pascal VOC PSPNet, clean images | mIoU: 75.38 -> 74.39 | Hurt |
| use joint objective (Ours-Co) instead of separate (Ours-Sep) in AT (ablation) | SegOOD AT · Pascal VOC PSPNet, SegOOD attack | mIoU: 30.15 -> 25.83 | Hurt |
| train AT with 3 iterations of SegOOD vs 7 iterations of baselines | SegOOD AT · Pascal VOC PSPNet, SegOOD attack | mIoU: 14.93 -> 30.15 | Improved |
| SegOOD AT vs RPPGD3-AT under transfer SegOOD | SegOOD AT · Pascal VOC PSPNet->DeepLabv3 | mIoU: 10.41 -> 19.30 | Improved |
Closest papers by meaning
- Towards Reliable Evaluation and Fast Training of Robust Semantic Segmentation Models (ECCV 2024 · similarity 0.92)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.91)
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks (CVPR 2018 · similarity 0.90)
- Dynamic Divide-and-Conquer Adversarial Training for Robust Semantic Segmentation (ICCV 2021 · similarity 0.89)
- RP-PGD: Boosting Segmentation Robustness with a Region-and-Prototype Based Adversarial Attack (AAAI 2025 · similarity 0.88)
- CosPGD: an efficient white-box adversarial attack for pixel-wise prediction tasks (ICML 2024 · 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.
