Localizing Memorized Regions in Diffusion Models via Coordinate-Wise Curvature Differences
For privacy and copyright protection in generative AI, this work provides a more precise way to identify which parts of generated images are memorized from training data.
The paper proposes a method to localize memorized regions in images generated by diffusion models by analyzing coordinate-wise curvature differences, outperforming prior attention-based localization on Stable Diffusion.
Diffusion models can unintentionally memorize training samples, raising concerns about privacy and copyright. While recent methods can detect memorization, they often rely on global or model-specific signals and provide limited insight into where memorization appears within a generated image. We provide a geometric characterization of local memorization as a coordinate-wise variance collapse. However, such collapse can also arise from intrinsic data constraints rather than overfitting. To isolate overfitting-driven memorization, we propose curvature-difference methods that subtract the curvature of an underfitted baseline, either the unconditional model or a less-trained version of itself. We further derive a score-difference proxy that provides a geometric explanation for the widely used score-difference-based detection metric. Experiments on Stable Diffusion, evaluated against ground-truth memorization masks, show that our method outperforms the prior attention-based localization method. Code is available at https://github.com/Gwangho99/mem-curv-diff.