Research map / Image generation / Editing and style transfer
IMPRESS: Evaluating the Resilience of Imperceptible Perturbations Against Unauthorized Data Usage in Diffusion-Based Generative AI
Bochuan Cao, Changjiang Li, Ting Wang, Jinyuan Jia, Bo Li, Jinghui Chen
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 |
|---|---|---|---|
| Purify GLAZE-protected images with IMPRESS before fine-tuning | IMPRESS · Style mimicking | CLIP classifier acc: 42.5 -> 87.0 | Improved |
| Purify GLAZE-protected images with IMPRESS before fine-tuning | IMPRESS · Style mimicking | Diffusion classifier acc: 42.9 -> 82.3 | Improved |
| Use JPEG compression instead of IMPRESS as post-processing | JPG baseline · Style mimicking | CLIP classifier acc: 42.5 -> 64.9 | Improved |
| Add Gaussian noise to protected images | Noise baseline · Style mimicking | CLIP classifier acc: 42.5 -> 47.9 | No clear effect |
| Add Gaussian noise to protected images | Noise baseline · Malicious editing | PSNR: 14.24 -> 11.06 | Hurt |
| Purify PhotoGuard-protected images with IMPRESS | IMPRESS · Malicious editing | PSNR: 14.24 -> 15.61 | Improved |
| Purify PhotoGuard-protected images with IMPRESS | IMPRESS · Malicious editing | VIF-p: 0.11 -> 0.16 | Improved |
| Use JPEG instead of IMPRESS on malicious editing | JPG baseline · Malicious editing | SSIM: 0.52 vs 0.49 | Mixed |
| Add consistency regularization to GLAZE (adaptive protection, beta=40) (ablation) | GLAZE · Style mimicking | CLIP classifier acc after IMPRESS: 87.0 -> 81.9 | No clear effect |
| Increase adaptive consistency weight beta to 1000 (ablation) | Adaptive GLAZE · Style mimicking | CLIP classifier acc of protected: 40.8 -> 77.1 | Hurt |
| Use O-PGD to balance adaptive objectives (ablation) | Adaptive GLAZE · Style mimicking | CLIP classifier acc after IMPRESS: 81.9 -> 83.9 | No clear effect |
| Compare with low-pass filtering, resize+flip+JPG and robust training | IMPRESS · Style mimicking | CLIP classifier acc: 51.3 -> 87.0 | Improved |
Closest papers by meaning
- Can Protective Perturbation Safeguard Personal Data from Being Exploited by Stable Diffusion? (CVPR 2024 · similarity 0.91)
- Adversarial Perturbations Cannot Reliably Protect Artists From Generative AI (ICLR 2025 · similarity 0.89)
- Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models (CVPR 2025 · similarity 0.89)
- Toward effective protection against diffusion-based mimicry through score distillation (ICLR 2024 · similarity 0.88)
- Prompt-Agnostic Adversarial Perturbation for Customized Diffusion Models (NeurIPS 2024 · similarity 0.88)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.88)
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.
