Research map / Label-efficient and robust learning / Recognition and long-tail
Similarity Distribution Based Membership Inference Attack on Person Re-identification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 inter-sample similarity distributions instead of feature embeddings alone | Membership inference on Re-ID models · Market1501 and DukeMTMC | attack success rate: 80.1% -> 87.0%; 80.5% -> 88.7% (ResNet50) | Helped |
| Add the anchor selector module to weight image-pair similarities (ablation) | M_SD membership inference attack · Market1501 and DukeMTMC | attack success rate: 87.0% -> 87.3%; 88.7% -> 89.1% (ResNet50) | Helped |
| Add the anchor selector module to weight image-pair similarities (ablation) | M_SD membership inference attack · Market1501 and DukeMTMC, MobileNetV2 and Xception | attack success rate: 80.6% -> 81.2%; 81.4% -> 82.2%; 89.7% -> 90.1%; 90.6% -> 91.6% | Helped |
| Reduce the percentage of sampled reference anchors (ablation) | M_SD membership inference attack · Market1501 and DukeMTMC Re-ID models | attack success rate | Hurt |
| Use the anchor selector when the reference set is small (ablation) | Similarity-distribution membership inference · Market1501 and DukeMTMC Re-ID models | attack success rate: 4% | Helped |
| Use similarity-distribution inference instead of the strongest loss-based attack | Membership inference on CIFAR10 classifiers · CIFAR10, ResNet18 | attack success rate: 78.7% -> 79.0% | Helped |
| Use similarity-distribution inference instead of the strongest loss-based attack | Membership inference on CIFAR10 classifiers · CIFAR10, ResNet50, VGG19, and GoogLeNet | attack success rate: 69.1% -> 68.3%; 63.9% -> 63.6%; 67.8% -> 66.9% | Hurt |
| Use a user-level attack with fewer positive images per identity (ablation) | User-level membership inference · Market1501 and DukeMTMC | attack success rate: M_U_high, M_U_mid, M_U_low | Hurt |
Closest papers by meaning
- Learning to Attack Real-World Models for Person Re-identification via Virtual-Guided Meta-Learning (AAAI 2021 · similarity 0.91)
- Intra-Inter Camera Similarity for Unsupervised Person Re-Identification (CVPR 2021 · similarity 0.87)
- A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification (AAAI 2019 · similarity 0.87)
- Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification (CVPR 2020 · similarity 0.87)
- Point to Set Similarity Based Deep Feature Learning for Person Re-Identification (CVPR 2017 · similarity 0.87)
- On the Difficulty of Membership Inference Attacks (CVPR 2021 · 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.