Research map / Image generation / Editing and style transfer
CLIP2Protect: Protecting Facial Privacy Using Text-Guided Makeup via Adversarial Latent Search
Fahad Shamshad, Muzammal Naseer, Karthik Nandakumar
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 |
|---|---|---|---|
| Use latent-manifold search with text-guided makeup and identity-preserving regularization instead of AMT-GAN | Black-box face verification impersonation · CelebA-HQ and LADN-Dataset | PSR (%): 52.84 -> 64.90 | Improved |
| Use CLIP2Protect instead of TIP-IM | Face identification dodging · LFW | Rank-1-U PSR: 77.8 -> 82.2 | Improved |
| Use CLIP2Protect instead of TIP-IM | Face identification dodging · LFW | Rank-5-U PSR: 54.9 -> 58.5 | Improved |
| Use CLIP2Protect instead of TIP-IM | Face identification impersonation · LFW | Rank-1-T PSR: 19.7 -> 23.4 | Improved |
| Use CLIP2Protect instead of TIP-IM | Face identification impersonation · LFW | Rank-5-T PSR: 41.9 -> 51.05 | Improved |
| Add identity-preserving latent regularization (ablation) | CLIP2Protect · Latent-loss ablation | FID: 43.2 -> 27.8 | Improved |
| Add identity-preserving latent regularization (ablation) | CLIP2Protect · Latent-loss ablation | PSR (%): 93.6 -> 76.2 | Hurt |
| Decrease latent-loss weight from 0.5 to 0.0001 (ablation) | CLIP2Protect · Latent-loss ablation | PSR (%): 31.2 -> 90.0 | Improved |
| Decrease latent-loss weight from 0.5 to 0.0001 (ablation) | CLIP2Protect · Latent-loss ablation | FID: 11.6 -> 38.4 | Hurt |
| Add makeup text guidance (ablation) | CLIP2Protect · Face verification impersonation | visual naturalness | Improved |
| Restrict adversarial optimization to identity-preserving latent codes (ablation) | CLIP2Protect · Face protection | human-perceived identity preservation | Improved |
| Vary the makeup prompt across five styles (ablation) | CLIP2Protect · CelebA-HQ | PSR: 74.1, 77.3, 78.4, 78.7, 79.2; Std. 1.24 | No clear effect |
Later papers that compared with this method
2 later accepted papers report CLIP2Protect as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- DeepProtect: Proactive Face-Swapping Defense using Identity Blending and Attribute Distortion (CVPR 2026 · better on 1 of 1 shared result)
- Machine Pareidolia: Protecting Facial Image with Emotional Editing (AAAI 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DiffAM: Diffusion-based Adversarial Makeup Transfer for Facial Privacy Protection (CVPR 2024 · similarity 0.89)
- Enhancing Facial Privacy Protection via Weakening Diffusion Purification (CVPR 2025 · similarity 0.88)
- Transferable Facial Privacy Protection against Blind Face Restoration via Domain-Consistent Adversarial Obfuscation (ICML 2024 · similarity 0.87)
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.86)
- SimAC: A Simple Anti-Customization Method for Protecting Face Privacy against Text-to-Image Synthesis of Diffusion Models (CVPR 2024 · similarity 0.86)
- IDProtector: An Adversarial Noise Encoder to Protect Against ID-Preserving Image Generation (CVPR 2025 · 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.
