Research map / Adversarial robustness and security / Black-box and transfer attacks
SEGA: A Transferable Signed Ensemble Gaussian Black-Box Attack Against No-Reference Image Quality Assessment Models
Yujia Liu, Dingquan Li, Zhixuan Li, Tiejun Huang
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 Gaussian smoothing of source gradients (with ensembling) (ablation) | SEGA · CLIVE (target DBCNN) | SROCC: 0.7858 -> 0.5714 | Improved |
| add Gaussian smoothing of source gradients (single source) (ablation) | SEGA · CLIVE (target DBCNN) | SROCC: 0.8357 -> 0.6698 | Improved |
| add gradient ensembling over three source models (with smoothing) (ablation) | SEGA · CLIVE (target DBCNN) | SROCC: 0.6698 -> 0.5714 | Improved |
| add gradient ensembling (no smoothing) (ablation) | SEGA · CLIVE (target DBCNN) | SROCC: 0.8357 -> 0.7858 | Improved |
| add Gaussian smoothing and ensembling together (ablation) | SEGA · CLIVE (target DBCNN) | MAE: 7.0034 -> 10.3865 | Improved |
| add both perturbation filter masks (ablation) | SEGA · CLIVE (target DBCNN) | SSIM: 0.749 -> 0.857 | Improved |
| add both perturbation filter masks (ablation) | SEGA · CLIVE (target DBCNN) | SROCC (transferability): 0.522 -> 0.562 | Hurt |
| add JND mask only (ablation) | SEGA · CLIVE (target DBCNN) | SSIM: 0.749 -> 0.808 | Improved |
| add filter mask M^F only (ablation) | SEGA · CLIVE (target DBCNN) | SSIM: 0.749 -> 0.817 | Improved |
| add both perturbation filter masks (ablation) | SEGA · CLIVE (target DBCNN) | LPIPS: 0.314 -> 0.235 | Improved |
| increase sigma from 5/255 to 10/255 (ablation) | SEGA smoothing sigma · CLIVE (target DBCNN) | PLCC: 0.7 -> 0.6 | Improved |
| increase sampling number m to 20 (ablation) | SEGA sampling number m · CLIVE (target DBCNN) | KROCC: below 0.4 | Improved |
Closest papers by meaning
- Backdoor Attacks Against No-Reference Image Quality Assessment Models via a Scalable Trigger (AAAI 2025 · similarity 0.90)
- Defense Against Adversarial Attacks on No-Reference Image Quality Models with Gradient Norm Regularization (CVPR 2024 · similarity 0.89)
- Guardians of Image Quality: Benchmarking Defenses Against Adversarial Attacks on Image Quality Metrics (ICML 2025 · similarity 0.85)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.83)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.83)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.83)
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.
