Research map / Adversarial robustness and security / Attacks on text and graphs
IAP: Invisible Adversarial Patch Attack through Perceptibility-Aware Localization and Perturbation Optimization
Subrat Kishore Dutta, Xiao 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 |
|---|---|---|---|
| increase patch size from 2% to 14% of the image (ablation) | IAP · ImageNet, Swin-B | attack success rate (%): 72.2% -> 99.4% | Improved |
| use patches of 4% or more coverage (ablation) | IAP · ImageNet, Swin-B | attack success rate (%): 90.7% or higher | Improved |
| increase regularization weight w3 moderately (ablation) | IAP · ImageNet, Swin-B | imperceptibility | Improved |
| increase regularization weight w3 excessively (ablation) | IAP · ImageNet, Swin-B | attack success and imperceptibility | Hurt |
| replace Adam update with the color-constant update rule (ablation) | IAP · ImageNet, Swin-B | imperceptibility | Improved |
| replace Adam update with the color-constant update rule (ablation) | IAP · ImageNet, Swin-B | attack success rate | Hurt |
| increase number of update iterations (ablation) | IAP · ImageNet, 6% patch | attack success rate | Mixed |
| use IAP instead of MPGD against SAC defense | IAP · ImageNet vs SAC, ResNet-50 | attack success rate (%): 11.6 -> 100 | Improved |
| use IAP instead of MPGD for imperceptibility | IAP · ImageNet, Swin-B | local LPIPS: 0.20 -> 0.07 | Improved |
| use IAP instead of MPGD for attack success | IAP · ImageNet, Swin-B | attack success rate (%): 70.50 -> 99.40 | Improved |
| use IAP instead of MPGD for human detectability | IAP · human study | detection rate (%): 94.5% -> 4.2% | Improved |
| use circular instead of square patch | IAP · ImageNet, Swin-B | attack success rate (%): 99.2% with LPIPS 0.085 | No clear effect |
Closest papers by meaning
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.91)
- False Positives Matter: Multidimensional Localization Evaluation and Training-Free Explainable Adversarial Patch Defense (AAAI 2026 · similarity 0.90)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.89)
- DAP: A Dynamic Adversarial Patch for Evading Person Detectors (CVPR 2024 · similarity 0.88)
- Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024 · similarity 0.88)
- PatchGuard: Adversarially Robust Anomaly Detection and Localization through Vision Transformers and Pseudo Anomalies (CVPR 2025 · similarity 0.88)
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.
