Research map / Adversarial robustness and security / Attacks on text and graphs
DDIM Inversion as a Perturbation Amplifier: Breaking Mimicry Protection via Reconstruction Error Minimization
Huming Qiu, Peiyi Chen, Mi Zhang, Geng Hong, Xiaoyu You, Mi Wen, Min Yang
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 Semantic Shortcut (SS) (ablation) | DIRP · DreamBooth, Metacloak | FID: 320.05 -> 212.71 | Improved |
| add Semantic Shortcut (SS) (ablation) | DIRP · DreamBooth, DIAGNOSIS | WACC: 100.00% -> 30.80% | Improved |
| add Bidirectional Asynchronous Optimization (BAC) (ablation) | DIRP · DreamBooth, DIAGNOSIS | WACC: 45.40% -> 30.80% | Improved |
| add BAC (ablation) | DIRP · DreamBooth, Metacloak | FID: 221.34 -> 212.71 | Improved |
| add Latent Residual Refinement (LRR) (ablation) | DIRP · DreamBooth, Metacloak | CS: 0.720 -> 0.828 | Improved |
| add LRR (ablation) | DIRP · DreamBooth, DIAGNOSIS | WACC: 81.00% -> 30.80% | Improved |
| add BAC (ablation) | DIRP · DreamBooth, DIAGNOSIS | FID: 126.12 -> 127.46 | No clear effect |
| apply DIRP purification | Flux.1-dev · Flux.1-dev, DIAGNOSIS | WACC: 99.90% -> 0.80% | Improved |
| apply DIRP purification | Flux.1-dev · Flux.1-dev, Metacloak | FID: 218.51 -> 195.08 | Improved |
| increase training epochs from 100 to 700 (ablation) | DIRP · Metacloak | FID: 217.67 (100), 200.93 (300), 212.71 (500), 214.70 (700) | Mixed |
| increase perturbation budget from 8/255 to 16/255 (ablation) | DIRP · Metacloak | FID: 236.40 -> 217.98 | Improved |
| increase perturbation budget from 24/255 to 32/255 (ablation) | DIRP · Metacloak | CS: 0.828 -> 0.802 | Hurt |
Closest papers by meaning
- Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models (CVPR 2025 · similarity 0.88)
- Toward effective protection against diffusion-based mimicry through score distillation (ICLR 2024 · similarity 0.87)
- StyleGuard: Preventing Text-to-Image-Model-based Style Mimicry Attacks by Style Perturbations (NeurIPS 2025 · similarity 0.86)
- Adversarial Perturbations Cannot Reliably Protect Artists From Generative AI (ICLR 2025 · similarity 0.86)
- MimicDiffusion: Purifying Adversarial Perturbation via Mimicking Clean Diffusion Model (CVPR 2024 · similarity 0.85)
- FakeInversion: Learning to Detect Images from Unseen Text-to-Image Models by Inverting Stable Diffusion (CVPR 2024 · similarity 0.85)
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.
