Research map / Adversarial robustness and security / Attacks on text and graphs
Universal Image Immunization against Diffusion-based Image Editing via Semantic Injection
Chanhui Lee, Donggyu Choi, Seunghyun Shin, Hae-Gon Jeon, Jeany Son
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 real training data to target injection loss (ablation) | Universal immunization UAP · Generated edit set, SD1.5 white-box | LPIPS: 0.557 -> 0.585 | Improved |
| add source semantic suppression loss (ablation) | Universal immunization UAP · Generated edit set, SD1.5 white-box | LPIPS: 0.585 -> 0.606 | Improved |
| add source semantic suppression loss (ablation) | Universal immunization UAP · Generated edit set, SD1.5 white-box | PSNR: 14.41 -> 14.19 | Improved |
| add source semantic suppression loss (ablation) | Universal immunization UAP · InstructPix2Pix black-box | LPIPS: 0.514 -> 0.527 | Improved |
| add real data to injection loss (ablation) | Universal immunization UAP · InstructPix2Pix black-box | LPIPS: 0.464 -> 0.514 | Improved |
| use semantic injection UAP instead of universalized Map baseline | Universal immunization · Generated edit set, SD1.5 white-box | LPIPS: 0.465 -> 0.606 | Improved |
| use semantic injection UAP instead of universalized Encoder baseline | Universal immunization · Generated edit set, SD1.4 black-box | PSNR: 18.08 -> 14.23 | Improved |
| universal UAP versus image-specific SA | Ours (universal) · Generated edit set, SD1.5 white-box | LPIPS: 0.490 -> 0.606 | Improved |
| universal UAP versus image-specific EA on Feat. Sim. | Ours (universal) · Generated edit set, SD1.5 white-box | Feat. Sim. (CLIP): 0.677 -> 0.673 | Hurt |
| universal UAP versus FastProtect optimization-free | Ours (universal) · ImageNet-Edit, SD1.5 white-box | LPIPS: 0.439 -> 0.609 | Improved |
| apply JPEG/diffusion purification to immunized images | Ours (universal) · Conditional DiffPure | LPIPS: 0.606 -> 0.568 | Mixed |
| vary the target concept (Ronaldo, Tiger, Sunflower, Peacock, Mandala) (ablation) | Universal immunization · Generated edit set, SD1.5 white-box | LPIPS: 0.602 avg, 0.015 std | No clear effect |
Closest papers by meaning
- Distraction is All You Need: Memory-Efficient Image Immunization against Diffusion-Based Image Editing (CVPR 2024 · similarity 0.92)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.89)
- AdvPaint: Protecting Images from Inpainting Manipulation via Adversarial Attention Disruption (ICLR 2025 · similarity 0.88)
- Pixel Is Not a Barrier: An Effective Evasion Attack for Pixel-Domain Diffusion Models (AAAI 2025 · similarity 0.87)
- AdvI2I: Adversarial Image Attack on Image-to-Image Diffusion Models (ICML 2025 · similarity 0.87)
- Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior (CVPR 2025 · similarity 0.87)
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.
