Research map / Adversarial robustness and security / Attacks on text and graphs
Towards Robust Defense against Customization via Protective Perturbation Resistant to Diffusion-based Purification
Wenkui Yang, Jie Cao, Junxian Duan, Ran He
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 patch-wise frequency guidance to the DDPM loss (ablation) | AntiPure · VGGFace2 (DreamBooth after purification) | ISM: 0.5958 -> 0.5644 | Improved |
| add patch-wise frequency guidance to the DDPM loss (ablation) | AntiPure · CelebA-HQ (DreamBooth after purification) | ISM: 0.6293 -> 0.6253 | No clear effect |
| add erroneous timestep guidance to the DDPM loss (ablation) | AntiPure · VGGFace2 (DreamBooth after purification) | ISM: 0.5958 -> 0.5901 | Improved |
| add erroneous timestep guidance to the DDPM loss (ablation) | AntiPure · CelebA-HQ (DreamBooth after purification) | ISM: 0.6293 -> 0.6489 | Hurt |
| use both guidances together (full AntiPure) over DDPM loss alone (ablation) | AntiPure · CelebA-HQ (DreamBooth after purification) | FID: 69.06 -> 81.15 | Improved |
| use both guidances together (full AntiPure) over DDPM loss alone (ablation) | AntiPure · VGGFace2 (DreamBooth after purification) | FID: 76.32 -> 90.77 | Improved |
| more purification iterations (10 to 40) against AntiPure (ablation) | AntiPure · CelebA-HQ (DreamBooth, purification iterations) | ISM: 0.6362 -> 0.5994 | Improved |
| more purification iterations (10 to 40) against Anti-DB (ablation) | Anti-DB · CelebA-HQ (DreamBooth, purification iterations) | ISM: 0.6020 -> 0.6391 | Hurt |
| replace Anti-DB with AntiPure after purification | DreamBooth P-C workflow · CelebA-HQ (DreamBooth after purification) | ISM: 0.6422 -> 0.6112 | Improved |
| swap DreamBooth for LoRA fine-tuning with AntiPure | AntiPure vs Anti-DB · VGGFace2 (LoRA after purification) | ISM: 0.5723 -> 0.5428 | Improved |
| lambda1/lambda2 0.25/0.75 instead of 0.5/0.5 (ablation) | AntiPure · VGGFace2 (DreamBooth after purification) | ISM: 0.5475 -> 0.5431 | No clear effect |
| lambda1/lambda2 0.75/0.25 instead of 0.5/0.5 (ablation) | AntiPure · VGGFace2 (DreamBooth after purification) | ISM: 0.5475 -> 0.5586 | No clear effect |
Closest papers by meaning
- Targeted Attack Improves Protection against Unauthorized Diffusion Customization (ICLR 2025 · similarity 0.90)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.90)
- Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models (CVPR 2025 · similarity 0.90)
- Harnessing Global-Local Collaborative Adversarial Perturbation for Anti-Customization (CVPR 2025 · similarity 0.89)
- SimAC: A Simple Anti-Customization Method for Protecting Face Privacy against Text-to-Image Synthesis of Diffusion Models (CVPR 2024 · similarity 0.89)
- Can Protective Perturbation Safeguard Personal Data from Being Exploited by Stable Diffusion? (CVPR 2024 · similarity 0.89)
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.
