Research map / Image generation / Editing and style transfer
Divide and Conquer: a Two-Step Method for High Quality Face De-identification with Model Explainability
Yunqian Wen, Bo Liu, Jingyi Cao, Rong Xie, Li Song
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 |
|---|---|---|---|
| separate privacy protection (3D ID code perturbation) from utility preservation (GAN restoration) into two steps | IDeudemon · CelebA-HQ | DIS and FID: DIS FR 0.819 vs 0.812 (best prior); FID 8.7 vs 22.6 | Improved |
| add visual similarity assistance (hybrid image with original hair and background) (ablation) | IDeudemon · FFHQ | FID / pose / expression: FID 25.78 -> 9.99 | Improved |
| add identity preserving loss (ablation) | IDeudemon GAN · FFHQ | DIS FR: 0.417 -> 0.804 | Improved |
| use cosine similarity instead of L1 distance in identity preserving loss (ablation) | IDeudemon GAN · FFHQ | DIS FR: 0.768 -> 0.804 | Improved |
| add reconstruction loss (ablation) | IDeudemon GAN · FFHQ | FID: 10.07 -> 9.99 | No clear effect |
| use the chosen GAN restoration model instead of CodeFormer, BOPB, GPEN, RestoreFormer or VQFR (ablation) | Step II restoration · FFHQ | FID: CodeFormer 10.83 -> 9.99 | Improved |
| increase protective noise scale above 0.3 (ablation) | ID code perturbation · FFHQ | visual quality of geometry | Hurt |
| use noise scale below 0.2 (ablation) | ID code perturbation · FFHQ | identity protection | Hurt |
Closest papers by meaning
- Personalized and Invertible Face De-Identification by Disentangled Identity Information Manipulation (ICCV 2021 · similarity 0.89)
- How Private is Private? A Comparative Study for Face De-Identification (NeurIPS 2026 · similarity 0.89)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.88)
- Protecting Facial Privacy: Generating Adversarial Identity Masks via Style-Robust Makeup Transfer (CVPR 2022 · similarity 0.88)
- Towards Open-Set Identity Preserving Face Synthesis (CVPR 2018 · similarity 0.88)
- DiffAM: Diffusion-based Adversarial Makeup Transfer for Facial Privacy Protection (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.
