Research map / Adversarial robustness and security / Attacks on text and graphs
SegPGD: An Effective and Efficient Adversarial Attack for Evaluating and Boosting Segmentation Robustness
Jindong Gu, Hengshuang Zhao, Volker Tresp, Philip H. S. Torr
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 |
|---|---|---|---|
| use SegPGD instead of PGD to create adversarial examples for 3-step adversarial training | PSPNet AT · Cityscapes PGD100 | mIoU: 3.95 -> 13.04 | Improved |
| use SegPGD instead of PGD for 3-step adversarial training | DeepLabv3 AT · VOC PGD100 | mIoU: 16.91 -> 18.24 | Improved |
| use SegPGD instead of PGD for 7-step adversarial training | DeepLabv3 AT · VOC PGD100 | mIoU: 19.10 -> 23.09 | Improved |
| use SegPGD instead of PGD for 7-step adversarial training | PSPNet AT · VOC PGD100 | mIoU: 16.97 -> 19.23 | Improved |
| use SegPGD instead of PGD for 7-step adversarial training | PSPNet AT · Cityscapes PGD100 | mIoU: 22.8 -> 23.13 | Improved |
| use SegPGD instead of PGD for 3-step adversarial training | DeepLabv3 AT · VOC PGD10 | mIoU: 28.16 -> 26.29 | Hurt |
| use SegPGD instead of PGD for 3-step adversarial training | PSPNet AT · VOC BIMl2 | mIoU: 51.56 -> 50.17 | Hurt |
| use SegFGSM instead of FGSM as the attack | standard PSPNet · VOC | mIoU: 36.76 -> 30.80 | Improved |
| use SegFGSM instead of FGSM as the attack | adversarially trained DeepLabV3 · VOC | mIoU: 46.78 -> 43.88 | Improved |
| use SegPGD instead of PGD as the attack for evaluating robustness (100 iterations) | PGD3-AT PSPNet · Cityscapes | mIoU: 3.95 -> 1.91 | Improved |
| add dynamic linear weighting schedule for misclassified pixels to PGD (ablation) | PGD attack · Standard PSPNet on VOC | mIoU | Improved |
| set constant lambda=0 (only-correct, as in DAG) instead of dynamic schedule (ablation) | SegPGD · VOC | mIoU | Hurt |
Closest papers by meaning
- RP-PGD: Boosting Segmentation Robustness with a Region-and-Prototype Based Adversarial Attack (AAAI 2025 · similarity 0.89)
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks (CVPR 2018 · similarity 0.88)
- Towards Reliable Evaluation and Fast Training of Robust Semantic Segmentation Models (ECCV 2024 · similarity 0.88)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.88)
- CosPGD: an efficient white-box adversarial attack for pixel-wise prediction tasks (ICML 2024 · similarity 0.87)
- ASAM: Boosting Segment Anything Model with Adversarial Tuning (CVPR 2024 · similarity 0.85)
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.
