Research map / Adversarial robustness and security / Attacks on text and graphs
VIP: Unified Certified Detection and Recovery for Patch Attack with Vision Transformers
Junbo Li, Huan Zhang, Cihang Xie
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 DeiT instead of original ViT as base classifier | ViP · ImageNet detection 2x2 | certified robustness: 72.0 -> 75.0 | Improved |
| use MAE pretraining instead of DeiT | ViP · ImageNet detection 2x2 | certified robustness: 75.0 -> 77.7 | Improved |
| use linear instead of quadratic mask sliding | ViP MAE base · ImageNet detection 2x2 | certified robustness: 77.7 -> 74.5 | Hurt |
| use MAE instead of ViT for dual-patch detection | ViP · ImageNet dual-patch 2x2 | certified robustness: 22.2 -> 42.0 | Improved |
| use MAE instead of DeiT for dual-patch detection | ViP · ImageNet dual-patch 4x4 | certified robustness: 16.6 -> 23.9 | Improved |
| replace CNN method with ViP MAE for recovery | ViP MAE base · ImageNet recovery 1% pixels | certified robustness: 32.3 -> 45.0 | Improved |
| use MAE instead of DeiT for certified recovery | ViP · ImageNet recovery 24x24 | adversarial robustness: 43.8 -> 45.0 | Improved |
| finetune 60 epochs instead of 30 (ablation) | ViP MAE base · ImageNet recovery 24x24 | adversarial robustness: 45.0 -> 46.0 | Improved |
| use stride 1 instead of 10 (ablation) | ViP MAE base · ImageNet recovery 24x24 | adversarial robustness: 38.8 -> 45.0 | Improved |
| scale from MAE base to MAE large (ablation) | ViP · ImageNet recovery 24x24 | adversarial robustness: 42.5 -> 48.3 | Improved |
| widen unmasked area from 19 to 37 (ablation) | ViP MAE base · ImageNet recovery 24x24 | adversarial robustness: 45.0 -> 45.1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ScaleCert — ScaleCert: Scalable Certified Defense against Adversarial Patches with Sparse Superficial Layers (NeurIPS 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Practical Certifiable Patch Defense With Vision Transformer (CVPR 2022 · similarity 0.92)
- Certified Defenses for Adversarial Patches (ICLR 2020 · similarity 0.91)
- Understanding and Defending Patched-based Adversarial Attacks for Vision Transformer (ICML 2023 · similarity 0.90)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.86)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · 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.
