Research map / Adversarial robustness and security / Attacks on text and graphs
Privacy-Preserving Adversarial Facial Features
Zhibo Wang, He Wang, Shuaifan Jin, Wenwen Zhang, Jiahui Hu, Yan Wang, Peng Sun, Wei Yuan, Kaixin Liu, Kui Ren
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 |
|---|---|---|---|
| Replace unprotected features with AdvFace adversarial features | Face recognition features · LFW | SRRA: 97.40% -> 4.03% | Improved |
| Replace unprotected features with AdvFace adversarial features | Face recognition features · CFP-FP | SRRA: 89.71% -> 18.43% | Improved |
| Replace unprotected features with AdvFace adversarial features | Face recognition features · AgeDB-30 | SRRA: 84.53% -> 16.67% | Improved |
| Add bounded adversarial latent noise through AdvFace | Face recognition features · LFW | accuracy: 98.13% -> 96.43% | Hurt |
| Add bounded adversarial latent noise through AdvFace | Face recognition features · CFP-FP | accuracy: 93.16% -> 90.59% | Hurt |
| Add bounded adversarial latent noise through AdvFace | Face recognition features · AgeDB-30 | accuracy: 87.57% -> 85.10% | Hurt |
| Retrain the server-side recognition network using adversarial features | AdvFace · LFW | accuracy: 96.43% -> 97.78% | Improved |
| Retrain the server-side recognition network using adversarial features | AdvFace · CFP-FP | accuracy: 90.59% -> 92.04% | Improved |
| Retrain the server-side recognition network using adversarial features | AdvFace · AgeDB-30 | accuracy: 85.10% -> 86.35% | Improved |
| Use TransRec rather than ResRec as the shadow model (ablation) | AdvFace · LFW, CFP-FP, AgeDB-30 | accuracy: 96.43% -> 96.53%; 90.59% -> 90.77%; 85.10% -> 85.13% | No clear effect |
| Use URec rather than ResRec as the shadow model (ablation) | AdvFace · LFW, CFP-FP, AgeDB-30 | accuracy: 96.43% -> 96.42%; 90.59% -> 90.13%; 85.10% -> 84.58% | Hurt |
Closest papers by meaning
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.90)
- Transferable Facial Privacy Protection against Blind Face Restoration via Domain-Consistent Adversarial Obfuscation (ICML 2024 · similarity 0.90)
- Enhancing Facial Privacy Protection via Weakening Diffusion Purification (CVPR 2025 · similarity 0.88)
- Asymmetric Invertible Threat: Learning Reversible Privacy Defense for Face Recognition (NeurIPS 2026 · similarity 0.88)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.88)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · 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.
