Research map / Federated learning and privacy / Privacy attacks and secure learning
Password-conditioned Anonymization and Deanonymization with Face Identity Transformers
Xiuye Gu, Weixin Luo, Michael S. Ryoo, Yong Jae Lee
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 feature dissimilarity loss L_dis (ablation) | face identity transformer · CASIA | identity multimodality (qualitative) | Improved |
| add explicit wrong-reconstruction training (WR) (ablation) | face identity transformer · CASIA | identity concealment under wrong password (qualitative) | Improved |
| add auxiliary password-predicting network Q with L_aux (ablation) | face identity transformer · CASIA | photo-realism and multimodality (qualitative) | Improved |
| add face classification loss on reconstruction L_rec_cls (ablation) | face identity transformer · CASIA | reconstruction quality (qualitative) | Improved |
| use adversarial face classification training instead of fixed non-adversarial classifier loss (ablation) | face identity transformer · CASIA | photo-realism (qualitative) | Improved |
| train identity change and recovery jointly end-to-end instead of separate deanonymizers for baselines (ablation) | face identity transformer · CASIA | LPIPS reconstruction error: 0.08 -> 0.03 | Improved |
| use face identity transformer instead of Blur baseline | face identity transformer · CASIA | LPIPS reconstruction error: 0.30 -> 0.03 | Improved |
| use face identity transformer instead of Ren et al | face identity transformer · LFW | L1 reconstruction error (LFW): 0.06 -> 0.04 | Improved |
| use wrong-password recovery instead of anonymization alone to hide identity | face identity transformer · CASIA AMT | percent judged same as input: 4.7 -> 0.7 | Improved |
| use 4 bits chunked N/4 16-way classifiers for 16-bit password in Q | face identity transformer · CASIA | classifier design | No clear effect |
Closest papers by meaning
- FaceRSA: RSA-Aware Facial Identity Cryptography Framework (AAAI 2024 · similarity 0.89)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.85)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.85)
- PerceptAnon: Exploring the Human Perception of Image Anonymization Beyond Pseudonymization for GDPR (ICML 2024 · similarity 0.84)
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.84)
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 2025 · 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.
