Research map / Adversarial robustness and security / Attacks on text and graphs
Breaking Rigidity in Adversarial Patch Attacks
Vishesh Kumar, Guha Balakrishnan, Akshay Agarwal
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 smooth deformable patch shape instead of circle (universal targeted attack) | SmuDPatch · ImageNet-Patch, ResNet50 | ASR: 27.9 -> 62.1 | Improved |
| use universal evolutionary-shape patch instead of per-image DAPatch | SmuDPatch · ImageNet-Patch, ResNet50 | ASR: 3.5 -> 62.1 | Improved |
| transfer deformable patches to unseen models (ViT, ConvNeXt-L) | SmuDPatch · ImageNet-Patch transfer | robust accuracy: ConvNeXt-L 83.6 -> 76.0; ViT 80.9 -> 65.5 | Mixed |
| attack adversarially robust models with deformable patches | adv-robust ResNet (Wong et al.) · ImageNet-Patch | robust accuracy: 53.5 -> 38.4 | Hurt |
| print patches and apply to real objects | ResNet50 · Physical-world (120 images) | accuracy: 66.6% -> 2.5% | Hurt |
| switch detector evaluation from square to deformable patches | NAPGuard · Patch detection | mAP: 64.07% -> 45.97% | Hurt |
| switch from square to deformable patches for VLM detection | LLaVA-NeXT · Patch detection (2000 images) | detections: 1,704 -> 765 | Hurt |
| switch from square to deformable patches for segmentation | SAM 3 · Patch segmentation | mIoU: 39.54 -> 9.41 | Hurt |
| apply SmuDPatch to object detectors | Faster R-CNN and YOLO · COCO detection | mAP: 41.86 -> 36.63 | Hurt |
Closest papers by meaning
- Certified Defenses for Adversarial Patches (ICLR 2020 · similarity 0.92)
- Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights (ICCV 2025 · similarity 0.91)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.90)
- Shape Matters: Deformable Patch Attack (ECCV 2022 · similarity 0.90)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · 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.
