Research map / Adversarial robustness and security / Black-box and transfer attacks
A Generative Victim Model for Segmentation
Aixuan Li, Jing Zhang, Jiawei Shi, Zhexiong Wan, Yiran Zhong, Yuchao Dai
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 |
|---|---|---|---|
| add diffusion-style noise and clip together (ablation) | score-based attack · COD10K ablation (R50 victim, query-based) | CC: .450 -> .323 | Improved |
| add noise on top of clip-only variant (ablation) | score-based attack · COD10K ablation (R50 victim, query-based) | CC: .450 -> .323 | Improved |
| add clip on top of noise-only variant (ablation) | score-based attack · COD10K ablation (R50 victim, query-based) | CC: .463 -> .323 | Improved |
| use score-derived perturbation instead of random Gaussian noise (ablation) | score-based attack · COD10K ablation (R50 victim, query-based) | CC: .653 -> .323 | Improved |
| vary omega from 30 to 90 (ablation) | score-based attack · COD10K ablation (R50 victim, query-based) | CC: .339 -> .323 | No clear effect |
| vary m_max from 20 to 30 steps (ablation) | score-based attack without query · COD10K (R50 victim) | CC: .377 -> .364 | Improved |
| increase m_max from 30 to 70 steps (ablation) | score-based attack without query · COD10K (R50 victim) | CC: .364 -> .401 | Hurt |
| add 100 queries to the victim model (ablation) | score-based attack · COD10K (R50 victim) | CC: .364 -> .323 | Improved |
| add 100 queries to the victim model (ablation) | score-based attack · SOD DUTS (R50 victim) | Corr: .716 -> .602 | Improved |
| use score-based attack versus MI-FGSM on robust victim | score-based attack · COD10K robust R50 victim (PVTv2 surrogate) | CC: .543 -> .483 | Improved |
| use score-based attack versus NAA on multi-class segmentation | score-based attack · VOC semantic segmentation transfer attack (PSPR50 surrogate, PSPR50 victim) | mIoU: 0.076 -> 0.520 | Hurt |
Closest papers by meaning
- Generative Adversarial Perturbations (CVPR 2018 · similarity 0.91)
- Adversarial Examples for Semantic Segmentation and Object Detection (ICCV 2017 · similarity 0.88)
- GAMA: Generative Adversarial Multi-Object Scene Attacks (NeurIPS 2022 · similarity 0.88)
- Evaluating the Adversarial Robustness of Semantic Segmentation: Trying Harder Pays Off (ECCV 2024 · similarity 0.87)
- RP-PGD: Boosting Segmentation Robustness with a Region-and-Prototype Based Adversarial Attack (AAAI 2025 · similarity 0.86)
- Image Hijacks: Adversarial Images can Control Generative Models at Runtime (ICML 2024 · similarity 0.84)
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.
