Research map / Adversarial robustness and security / Certified and empirical robustness
CertMask: Certifiable Defense Against Adversarial Patches via Theoretically Optimal Mask Coverage
Xuntao Lyu, Ching-Chi Lin, Abdullah Al Arafat, Georg von der Brüggen, Jian-Jia Chen, Zhishan Guo
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 CertMask instead of PatchCleanser | ViT · ImageNet 2% patch | certified robust accuracy: 62.1 -> 75.5 | Improved |
| use CertMask instead of PatchCleanser | ViT · CIFAR-10 2.4% patch | certified robust accuracy: 89.1 -> 96.4 | Improved |
| use CertMask instead of PatchCleanser | ViT · ImageNette 2% patch | certified robust accuracy: 96.4 -> 99.0 | Improved |
| use CertMask instead of PatchCleanser | ResNet · ImageNette 2% patch | certified robust accuracy: 94.4 -> 98.9 | Improved |
| use CertMask instead of PG-DS | DS · ImageNet 2% patch | certified robust accuracy: 15.7 -> 75.5 | Improved |
| use CertMask instead of PatchCleanser | ViT · ImageNet 2% patch | clean accuracy: 83.9 -> 84.5 | No clear effect |
| use Offset Tiling instead of Replicated Tiling (ablation) | ViT · ImageNet 3% patch | certified robust accuracy: 78.6 -> 74.0 | Hurt |
| use Offset Tiling instead of Replicated Tiling (ablation) | ResNet · ImageNet 3% patch | certified robust accuracy: 73.7 -> 68.5 | Hurt |
| use Offset Tiling instead of Replicated Tiling (ablation) | MLP · CIFAR-10 2.4% patch | clean accuracy: 95.7 -> 98.0 | Improved |
| increase coverage k from 1 to 6 (ablation) | ViT · ImageNet 3% patch | certified robust accuracy: 78.6 -> 74.0 | Hurt |
| increase coverage k from 1 to 6 (ablation) | ViT · ImageNet 3% patch | clean accuracy: 82.0 -> 84.5 | Improved |
| increase mask size from 40px to 120px (ablation) | ViT · ImageNet (k=6) | certified robust accuracy: 76 -> 62 | Hurt |
Closest papers by meaning
- (De)Randomized Smoothing for Certifiable Defense against Patch Attacks (NeurIPS 2020 · similarity 0.90)
- Certified Defenses for Adversarial Patches (ICLR 2020 · similarity 0.90)
- VIP: Unified Certified Detection and Recovery for Patch Attack with Vision Transformers (ECCV 2022 · similarity 0.89)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.87)
- Et Tu Certifications: Robustness Certificates Yield Better Adversarial Examples (ICML 2024 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
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.
