Research map / Adversarial robustness and security / Attacks on text and graphs
Universal Perturbation Attack Against Image Retrieval
Jie Li, Rongrong Ji, Hong Liu, Xiaopeng Hong, Yue Gao, Qi Tian
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 pair-wise ranking relationship instead of label-wise classification-style loss | UAP against retrieval · V-MAC retrieval | mean dropping rate (mDR): 61.80% -> 66.94% | Improved |
| use pair-wise instead of label-wise loss | UAP against retrieval · R-GeM retrieval | mean dropping rate (mDR): 23.76% -> 66.69% | Improved |
| add list-wise (NDCG-based) relationship on top of pair-wise (ablation) | UAP against retrieval · A-MAC retrieval | mean dropping rate (mDR): 44.35% -> 48.33% | Improved |
| add list-wise relationship on top of pair-wise (ablation) | UAP against retrieval · R-MAC retrieval | mean dropping rate (mDR): 59.71% -> 62.60% | Improved |
| add list-wise relationship on top of pair-wise (ablation) | UAP against retrieval · V-MAC retrieval | mean dropping rate (mDR): 66.94% -> 67.96% | Improved |
| use ranking-distilled substitute model instead of ImageNet pre-trained model for black-box attack (ablation) | UAP against retrieval · A-GeM | mean dropping rate (mDR): 32.98% -> 39.72% | Improved |
| use ranking-distilled substitute model instead of ImageNet pre-trained model (ablation) | UAP against retrieval · V-GeM | mean dropping rate (mDR): 28.85% -> 44.68% | Improved |
| use ranking-distilled model instead of randomly initialized model (ablation) | UAP against retrieval · A-GeM | mean dropping rate (mDR): 5.53% -> 39.72% | Improved |
| use multi-scale random resizing [256,1024] instead of fixed 362x362 (ablation) | list-wise UAP · A-GeM | mean dropping rate (mDR): 16.89% -> 56.88% | Improved |
| use multi-scale random resizing [256,1024] instead of fixed 1024x1024 (ablation) | list-wise UAP · V-GeM | mean dropping rate (mDR): 30.42% -> 63.72% | Improved |
| narrow the random resizing range from [256,1024] to [768,1024] (ablation) | list-wise UAP · V-GeM | mean dropping rate (mDR): 63.72% -> 42.08% | Hurt |
| widen the random resizing range from [256,1024] to [128,1024] (ablation) | list-wise UAP · A-GeM | mean dropping rate (mDR): 56.88% -> 53.21% | Hurt |
Closest papers by meaning
- QAIR: Practical Query-Efficient Black-Box Attacks for Image Retrieval (CVPR 2021 · similarity 0.89)
- Data-Free Universal Adversarial Perturbation and Black-Box Attack (ICCV 2021 · similarity 0.89)
- Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior (CVPR 2025 · similarity 0.88)
- Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models (AAAI 2025 · similarity 0.87)
- Once and for All: Universal Transferable Adversarial Perturbation against Deep Hashing-Based Facial Image Retrieval (AAAI 2024 · similarity 0.86)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.85)
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.
