Research map / Image generation / Text-to-image diffusion
Demystifying Foreground-Background Memorization in Diffusion Models
Jimmy Z. Di, Yiwei Lu, Yaoliang Yu, Gautam Kamath, Adam Dziedzic, Franziska Boenisch
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 |
|---|---|---|---|
| Aggregate neurons across a prompt cluster (NeMo-C) instead of per-prompt neurons | NeMo · 500 memorized prompts | mitigation score: 0.74 -> 0.83 | Improved |
| NeMo-C vs NeMo on image quality | NeMo · Image quality | Q-Align: 3.63 -> 3.52 | Hurt |
| NeMo-C vs NeMo on image quality | NeMo · Image quality | DB-CNN: 0.587 -> 0.586 | No clear effect |
| Apply any model-level mitigation (NeMo, Wanda, DetectMem) | Stable Diffusion v1.4 · Verbatim memorization | VM rate | Improved |
| Apply any model-level mitigation | Stable Diffusion v1.4 · Foreground memorization | FM rate | No clear effect |
| Apply mitigation methods | Stable Diffusion v1.4 · One-to-many correspondence | matched training images | No clear effect |
| Use MS-SSIM instead of SSCD in FB-Mem (ablation) | FB-Mem · TM vs NM classification | AUROC: 0.992 -> 0.962 | Hurt |
| Use MS-SSIM instead of SSIM for VM vs NM (ablation) | FB-Mem · VM vs NM classification | accuracy: 0.954 -> 0.997 | Improved |
Closest papers by meaning
- Exploring Local Memorization in Diffusion Models via Bright Ending Attention (ICLR 2025 · similarity 0.91)
- Detecting, Explaining, and Mitigating Memorization in Diffusion Models (ICLR 2024 · similarity 0.91)
- How Diffusion Models Memorize (NeurIPS 2026 · similarity 0.89)
- Towards Memorization-Free Diffusion Models (CVPR 2024 · similarity 0.88)
- On the Edge of Memorization in Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Understanding and Mitigating Memorization in Generative Models via Sharpness of Probability Landscapes (ICML 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.
