Research map / Adversarial robustness and security / Attacks on text and graphs
CamoPatch: An Evolutionary Strategy for Generating Camoflauged Adversarial Patches
Phoenix Neale Williams, Ke Li
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 |
|---|---|---|---|
| remove simulated annealing from location optimization (random search) (ablation) | CamoPatch · AT-WideResNet-50-2 | ASR: 87.02% -> 83.02% | Hurt |
| remove simulated annealing from location optimization (ablation) | CamoPatch · AT-ResNet-50 | ASR: 94.00% -> 90.00% | Hurt |
| remove simulated annealing from location optimization (ablation) | CamoPatch · VGG-16 | ASR: 90.30% -> 90.01 | No clear effect |
| remove simulated annealing from location optimization (ablation) | CamoPatch · ResNet-50 | ASR: 90.00% -> 90.00% | No clear effect |
| increase patch pattern optimization budget li from 1 to 4 (ablation) | CamoPatch · VGG-16 | accuracy after attack: 12.88% -> 9.64% | Improved |
| increase location exploration t from 100 to 300 (ablation) | CamoPatch · VGG-16 | accuracy after attack: 10.79% -> 9.64% | Improved |
| increase sigma from 0.1 to 0.3 (ablation) | CamoPatch · VGG-16 | accuracy after attack: 9.64% -> 11.66% | Hurt |
| increase number of circles N from 100 to 300 (ablation) | CamoPatch · VGG-16 | accuracy and runtime | Hurt |
| minimize l2 of patch in Patch-RS (Patch-RS*) | Patch-RS · ImageNet classifiers | patch l2 distance: 0.62 -> 0.42 | Improved |
| use circle-based RGB patch with arbitrary colors instead of corner-of-color-cube patch | CamoPatch vs Patch-RS · ImageNet classifiers | patch l2 distance: 0.71 -> 0.09 | Improved |
| adversarial training of classifier | AT-ResNet-50 · ImageNet | accuracy after CamoPatch attack: 10.00% -> 6.00% | Hurt |
| use WideResNet instead of ResNet in adversarial training | AT classifier · ImageNet | accuracy after CamoPatch attack: 6.00% -> 12.98% | Improved |
Closest papers by meaning
- Breaking Rigidity in Adversarial Patch Attacks (ECCV 2026 · similarity 0.89)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.88)
- Generating Natural Images with Direct Patch Distributions Matching (ECCV 2022 · similarity 0.88)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.88)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.86)
- DepthVanish: Optimizing Adversarial Interval Structures for Stereo-Depth-Invisible Patches (NeurIPS 2025 · 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.
