Research map / Image generation / Text-to-image diffusion
Targeted Data Protection for Diffusion Model by Matching Training Trajectory
Hojun Lee, Mijin Koo, Yeji Song, Nojun Kwak
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 |
|---|---|---|---|
| match the full training trajectory instead of snapshots | Mist/T-ASPL style targeted protection · Targeted identity protection | ISM w/ target: 0.147 -> 0.393 | Improved |
| match the full training trajectory instead of snapshots | T-ASPL · Targeted identity protection | ISM w/ data to protect: 0.226 -> 0.202 | Improved |
| trajectory matching vs ACE | ACE · Targeted identity protection | ISM w/ data to protect: 0.405 -> 0.202 | Improved |
| trajectory matching vs prior targeted losses | T-ASPL · Targeted identity protection | BRISQUE: 23.69 -> 11.51 | Improved |
| trajectory matching vs prior targeted losses | Mist · Targeted identity protection | SER-FIQ: 0.65 -> 0.80 | Improved |
| apply Gaussian blur kernel 3 to protected images (ablation) | TAFAP · Targeted identity protection | ISM w/ target: 0.393 -> 0.353 | No clear effect |
| apply Gaussian blur kernel 9 (ablation) | TAFAP · Targeted identity protection | ISM w/ data to protect: 0.202 -> 0.496 | Hurt |
| apply JPEG quality 10 (ablation) | TAFAP · Targeted identity protection | ISM w/ target: 0.393 -> 0.105 | Hurt |
| apply 0.5x downscaling then restoration (ablation) | TAFAP · Targeted identity protection | ISM w/ data to protect: 0.202 -> 0.496 | Hurt |
| apply 2x upscale then restoration (ablation) | TAFAP · Targeted identity protection | ISM w/ target: 0.393 -> 0.304 | Hurt |
| attacker uses SD1.4 without LoRA or SD2.1 (ablation) | TAFAP · Targeted identity protection | qualitative deviation from protected identity | Mixed |
| increase noise update iterations (ablation) | TAFAP · Noise update iterations | identity change | Improved |
Closest papers by meaning
- Perturb a Model, Not an Image: Towards Robust Privacy Protection via Anti-Personalized Diffusion Models (NeurIPS 2025 · similarity 0.87)
- Anti-Diffusion: Preventing Abuse of Modifications of Diffusion-Based Models (AAAI 2025 · similarity 0.86)
- Can Protective Perturbation Safeguard Personal Data from Being Exploited by Stable Diffusion? (CVPR 2024 · similarity 0.86)
- DiffusionGuard: A Robust Defense Against Malicious Diffusion-based Image Editing (ICLR 2025 · similarity 0.86)
- Membership Inference on Text-to-Image Diffusion Models via Conditional Likelihood Discrepancy (NeurIPS 2024 · similarity 0.86)
- DIAGNOSIS: Detecting Unauthorized Data Usages in Text-to-image Diffusion Models (ICLR 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.
