Research map / Image generation / Text-to-image diffusion
Anonymization Prompt Learning for Facial Privacy-Preserving Text-to-Image Generation
Liang Shi, Jie Zhang, Shiguang Shan
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 |
|---|---|---|---|
| prepend learned Anonymization Prompt | SD-v1.5 · Test identities | ID-ACC mean: 0.30254 -> 0.03970 | Improved |
| transfer prompt trained on SD-v1.5 to SDXL | SDXL · Test identities | ID-ACC mean: 0.27218 -> 0.07852 | Improved |
| transfer prompt to RealisticVision v5.1 | RealisticVision · Test identities | ID-ACC mean: 0.32071 -> 0.09033 | Improved |
| APL instead of Concept Ablation on facial attribute preservation | Concept Ablation · ID-specific generation | Attr-ACC: 0.53898 -> 0.60222 | Improved |
| APL instead of ESD on non-ID quality | ESD · COCO-30k | FID: 9.59 -> 2.98 | Improved |
| APL instead of Concept Ablation on ID-specific FID | Concept Ablation · ID-specific generation | FID: 57.58 -> 20.01 | Improved |
| apply prompt to Dreambooth-learned identities | SD-v1.5 Dreambooth · VGGFace-2 new identities | identity accuracy: 0.53 -> 0.43 | Improved |
| remove regularization loss L_Reg (ablation) | APL · Non-ID prompts | semantic alignment | Improved |
| prompt length 10 instead of 1 (ablation) | APL · Test identities | ID-ACC mean: 0.11434 -> 0.03257 | Improved |
| prompt length 20 instead of 10 (ablation) | APL · Test identities | ID-ACC mean: 0.03257 -> 0.04084 | No clear effect |
| alpha 1 instead of 0 (ablation) | APL · Test identities | ID-ACC mean: 0.05688 -> 0.03257 | Improved |
| alpha 2 instead of 1 (ablation) | APL · Test identities | ID-ACC mean: 0.03257 -> 0.06707 | Hurt |
Closest papers by meaning
- Facial Identity Anonymization via Intrinsic and Extrinsic Attention Distraction (CVPR 2024 · similarity 0.83)
- MYOPIA: Protecting Face Privacy from Malicious Personalized Text-to-Image Synthesis via Unlearnable Examples (AAAI 2025 · similarity 0.81)
- Enhancing Facial Privacy Protection via Weakening Diffusion Purification (CVPR 2025 · similarity 0.81)
- Balancing Privacy and Performance: A Many-in-One Approach for Image Anonymization (AAAI 2025 · similarity 0.81)
- Prompt-Agnostic Adversarial Perturbation for Customized Diffusion Models (NeurIPS 2024 · similarity 0.80)
- Enhancing Privacy-Utility Trade-offs to Mitigate Memorization in Diffusion Models (CVPR 2025 · similarity 0.80)
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.
