Research map / Adversarial robustness and security / Black-box and transfer attacks
Decision-based Black-box Attack Against Vision Transformers via Patch-wise Adversarial Removal
Yucheng Shi, Yahong Han, Yu-an Tan, Xiaohui Kuang
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 |
|---|---|---|---|
| initialize with PAR noise | HSJA · ILSVRC-2012 res-101 | median l2 distance: 35.13 -> 8.92 | Improved |
| initialize with PAR noise | Sign-OPT · ILSVRC-2012 res-101 | median l2 distance: 50.18 -> 9.16 | Improved |
| initialize with PAR noise | Boundary · ILSVRC-2012 res-101 | median l2 distance: 16.25 -> 7.29 | Improved |
| initialize with PAR noise | CAB · ILSVRC-2012 res-101 | median l2 distance: 11.20 -> 4.73 | Improved |
| initialize with PAR noise | BBA · ILSVRC-2012 res-101 | median l2 distance: 9.00 -> 4.89 | Improved |
| initialize with PAR noise | CAB · ILSVRC-2012 vit_s16 | average l2 distance: 44.85 -> 24.96 | Improved |
| initialize with PAR noise | Boundary · ILSVRC-2012 vit_s16 | median l2 distance: 21.52 -> 11.09 | Improved |
| initialize with PAR noise | HSJA · ILSVRC-2012 vit_s16 | median l2 distance: 65.81 -> 14.55 | Improved |
| PAR alone versus initial Gaussian noise | PAR · ILSVRC-2012 res-101 | median l2 distance: 58.60 -> 9.19 | Improved |
| lower minimum patch size (56 down to 7), initial patch 112 (ablation) | PAR + Boundary · ILSVRC-2012 vgg-19 | median noise: 6.21 -> 4.31 | Improved |
| smaller minimum patch size 7 vs 56 (ablation) | PAR + Boundary · ILSVRC-2012 vgg-19 | average query number: 16.80 -> 195.69 | Hurt |
| smaller initial patch size (112 to 14), min patch 7 (ablation) | PAR + Boundary · ILSVRC-2012 vgg-19 | average query number: 195.69 -> 415.06 | Hurt |
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.
- Sign-OPT — Sign-OPT: A Query-Efficient Hard-label Adversarial Attack (ICLR 2020) · better on 3 of 3 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
- Patch-Fool: Are Vision Transformers Always Robust Against Adversarial Perturbations? (ICLR 2022 · similarity 0.91)
- Towards Understanding and Improving Adversarial Robustness of Vision Transformers (CVPR 2024 · similarity 0.90)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.88)
- BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks (ICML 2024 · similarity 0.88)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.88)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.87)
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.
