Research map / Adversarial robustness and security / Black-box and transfer attacks
Bias-based Universal Adversarial Patch Attack for Automatic Check-out
Aishan Liu, Jiakai Wang, Xianglong Liu, Bowen Cao, Chongzhi Zhang, Hang Yu
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 patch with the fused textural prior instead of simple fusion of simple examples (ablation) | Bias-based patch · RPC white-box (ResNet-152) | top-1 accuracy after attack: 24.09 -> 5.42 | Improved |
| initialize patch with the fused textural prior instead of a white patch (ablation) | Bias-based patch · RPC white-box (ResNet-152) | top-1 accuracy after attack: 17.67 -> 5.42 | Improved |
| train with 1000 prototypes instead of 1000 item images (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 12.43 -> 6.51 | Improved |
| replace item images with prototypes in the training mix (750 : 250) (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 12.43 -> 7.81 | Improved |
| use 500 prototypes plus 500 item images instead of 1000 item images (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 12.43 -> 10.03 | Improved |
| use 250 prototypes plus 750 item images instead of 1000 item images (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 12.43 -> 11.55 | Improved |
| use 10000 item images instead of 1000 prototypes (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 6.51 -> 5.40 | Mixed |
| use 2000 item images instead of 1000 item images (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 12.43 -> 6.57 | Improved |
| add rotation transformation during training (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 13.01 -> 11.98 | Improved |
| add distortion transformation during training (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 30.70 -> 22.26 | Improved |
| add affine transformation during training (ablation) | Bias-based patch · RPC (ResNet-152) | top-1 accuracy after attack: 21.10 -> 16.38 | Improved |
| use proposed method vs best compared method on unseen classes | Bias-based patch · RPC unseen classes | top-1 accuracy after attack: 31.62 -> 7.23 | Improved |
Closest papers by meaning
- Perceptual-Sensitive GAN for Generating Adversarial Patches (AAAI 2019 · similarity 0.87)
- CamoPatch: An Evolutionary Strategy for Generating Camoflauged Adversarial Patches (NeurIPS 2023 · similarity 0.87)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.86)
- Pandora's Box: Towards Building Universal Attackers against Real-World Large Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.86)
- BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks (ICML 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.
