Research map / Adversarial robustness and security / Attacks on text and graphs
Shape Matters: Deformable Patch Attack
Zhaoyu Chen, Bo Li, Shuang Wu, Jianghe Xu, Shouhong Ding, Wenqiang Zhang
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 |
|---|---|---|---|
| optimize a deformable triangle-based shape instead of a fixed circle patch with white texture (ablation) | white patch · ILSVRC2012 shape-only, 3% area, ResNet-152 | ASR (%): 3.3 -> 23.6 | Improved |
| keep the deformed shape instead of using its larger convex hull (ablation) | DAPatch · ILSVRC2012 shape-only, 3% area, ResNet-152 | ASR (%): 5.0 -> 23.6 | Improved |
| add multiple anchors (3) to the deformable patch representation (ablation) | DAPatch · ILSVRC2012 untargeted, 1% area, ResNet-152 | ASR (%): 78.8 -> 82.3 | Improved |
| add multiple anchors (3) to the deformable patch representation (ablation) | DAPatch · ILSVRC2012 untargeted, 3% area, ResNet-152 | ASR (%): 97.9 -> 99.4 | Improved |
| add multiple anchors (3) to the deformable patch representation (ablation) | DAPatch · ILSVRC2012 untargeted, 1% area, ViT-B/16 | ASR (%): 80.9 -> 80.2 | No clear effect |
| use deformable patch instead of fixed circular patch (GAP c) | patch attack · ILSVRC2012 untargeted, 1% area, ResNet-152 | ASR (%): 74.4 -> 78.8 | Improved |
| use deformable patch instead of PS-GAN in targeted attack | patch attack · ILSVRC2012 targeted, 7% area, ResNet-152 | ASR (%): 87.4 -> 90.2 | Improved |
| use deformable patch instead of fixed shapes against EfficientNet-b7 targeted | patch attack · ILSVRC2012 targeted, 3% area, EfficientNet-b7 | ASR (%): 53.6 -> 53.6 | No clear effect |
| use deformable patch instead of GAP c against shape-texture debiased network | patch attack · ILSVRC2012 untargeted, 1% area, ResNet50-Debiased | ASR (%): 68.6 -> 68.6 | No clear effect |
| use deformable patch instead of LaVAN c on GTSRB | patch attack · GTSRB untargeted, 3% area, ResNet-152 | ASR (%): 58.4 -> 61.5 | Improved |
| use deformable patch instead of GAP c against LGS defense | patch attack · ILSVRC2012 2% area, LGS, ResNet-152 | ASR (%): 51.4 -> 53.9 | Improved |
| use deformable patch instead of PS-GAN in the physical world | patch attack · Physical attack, middle lighting, 0 degrees | ASR (%): 44 -> 60 | Improved |
Closest papers by meaning
- Breaking Rigidity in Adversarial Patch Attacks (ECCV 2026 · similarity 0.90)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.85)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.85)
- DepthVanish: Optimizing Adversarial Interval Structures for Stereo-Depth-Invisible Patches (NeurIPS 2025 · similarity 0.84)
- Hide in Thicket: Generating Imperceptible and Rational Adversarial Perturbations on 3D Point Clouds (CVPR 2024 · similarity 0.84)
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.
