Research map / Image generation / Editing and style transfer
DeepProtect: Proactive Face-Swapping Defense using Identity Blending and Attribute Distortion
Eungi Lee, Seung-hyeok Back, Hyung-Il Kim, Seok Bong Yoo
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 generator optimization with identity-lock loss to identity blending (ablation) | DeepProtect · CelebA-HQ, SimSwap | SSIM of protected image: 0.584 -> 0.915 | Improved |
| add generator optimization when combining identity blending and attribute distortion (ablation) | DeepProtect · CelebA-HQ, SimSwap | PDS: -0.196 -> 0.192 | Improved |
| add identity blending to attribute distortion (ablation) | DeepProtect · CelebA-HQ, SimSwap | ISM: 0.419 -> 0.201 | Improved |
| add attribute distortion to identity blending with generator optimization (ablation) | DeepProtect · CelebA-HQ, SimSwap | PDS: 0.136 -> 0.192 | Improved |
| use identity blending alone without generator optimization (ablation) | DeepProtect · CelebA-HQ, SimSwap | SSIM of protected image: 0.584 | Hurt |
| combine identity blending with attribute distortion instead of attribute distortion only (ablation) | DeepProtect · CelebA-HQ, SimSwap | DSR (%): 60.5 -> 94.8 | Improved |
| use DeepProtect Combined instead of FaceShield | DeepProtect · CelebA-HQ, SimSwap | DSR (%): 81.6 -> 94.8 | Improved |
| apply LoRA to the affine transform module and tune only middle layers | DeepProtect generator tuning · Complexity | trainable parameters: approximately 11M trainable parameters before LoRA | Improved |
| apply Gaussian blur post-processing to protected images | DeepProtect · CelebA-HQ, SimSwap | DSR (%): 94.8 -> 93.5 | No clear effect |
| apply adaptive purification attack | DeepProtect · CelebA-HQ, SimSwap | DSR (%): 94.8 -> 93.2 | No clear effect |
| noise-based defenses under imperceptibility constraint | CMUA-Watermark and DF-RAP · CelebA-HQ, SimSwap | DSR (%): 41.0 and 42.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CLIP2Protect — CLIP2Protect: Protecting Facial Privacy Using Text-Guided Makeup via Adversarial Latent Search (CVPR 2023) · 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
- NullSwap: Proactive Identity Cloaking Against Deepfake Face Swapping (ICCV 2025 · similarity 0.91)
- Implicit Identity Driven Deepfake Face Swapping Detection (CVPR 2023 · similarity 0.89)
- Guard Me If You Know Me: Protecting Specific Face-Identity from Deepfakes (NeurIPS 2025 · similarity 0.88)
- DeepFake Detection Based on Discrepancies Between Faces and Their Context (TPAMI 2021 · similarity 0.88)
- High-Fidelity Diffusion Face Swapping with ID-Constrained Facial Conditioning (CVPR 2026 · similarity 0.88)
- DiffusionFake: Enhancing Generalization in Deepfake Detection via Guided Stable Diffusion (NeurIPS 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.
