Research map / Adversarial robustness and security / Black-box and transfer attacks
SAM Encoder Breach by Adversarial Simplicial Complex Triggers Downstream Model Failures
Yi Qin, Rui Wang, Tao Huang, Tong Xiao, Liping Jing
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 domain re-adaptation (DRA) initialization to ASC (ablation) | VeSCA · ISTD | BER: 6.258 -> 11.517 | Improved |
| add domain re-adaptation constraint with BSR augmentation (ablation) | ASC+BSR · ISTD | BER: 6.012 -> 11.517 | Improved |
| add PAR augmentation on top of DRA (ablation) | VeSCA · ISTD | BER: 11.517 -> 13.228 | Improved |
| use PAR instead of BSR augmentation (ablation) | ASC · ISTD | BER: 6.012 -> 6.258 | Improved |
| use PAR instead of DTSI augmentation (ablation) | ASC · ISTD | BER: 5.231 -> 6.258 | Improved |
| use DRA instead of MUI-GRAT universal perturbation initialization (ablation) | ASC · ISTD | BER: 10.763 -> 11.517 | Improved |
| use VeSCA instead of MUI-GRAT | downstream attack · ISTD | BER: 11.960 -> 13.228 | Improved |
| use VeSCA instead of MUI-GRAT | downstream attack · CityScale | Recall %: 22.964 -> 20.045 | Improved |
| use VeSCA instead of MANDA | downstream attack · SegInW | MAP %: 32.263 -> 28.934 | Improved |
| use VeSCA instead of MUI-GRAT | downstream attack · CAMO | S_alpha: 0.335 -> 0.316 | Improved |
| use white-box Attack-SAM instead of encoder-only transfer attacks | Attack-SAM · ISTD | BER: 1.433 vs MIM 3.008 | Hurt |
| use more than 40 source-domain images in domain re-adaptation (ablation) | VeSCA · ISTD | BER | No clear effect |
Closest papers by meaning
- Transferable Adversarial Attacks on SAM and Its Downstream Models (NeurIPS 2024 · similarity 0.91)
- Robust SAM: On the Adversarial Robustness of Vision Foundation Models (AAAI 2025 · similarity 0.90)
- Vanish into Thin Air: Cross-prompt Universal Adversarial Attacks for SAM2 (NeurIPS 2025 · similarity 0.89)
- DarkSAM: Fooling Segment Anything Model to Segment Nothing (NeurIPS 2024 · similarity 0.87)
- ASAM: Boosting Segment Anything Model with Adversarial Tuning (CVPR 2024 · similarity 0.86)
- X-Transfer Attacks: Towards Super Transferable Adversarial Attacks on CLIP (ICML 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.
