Demystifying Foreground-Background Memorization in Diffusion Models

TL;DR

This paper introduces FB-Mem, a segmentation-based metric to quantify partial and complex memorization patterns in diffusion models, revealing limitations of current mitigation techniques and proposing a new clustering-based solution.

Contribution

It presents a novel segmentation-based metric for detailed memorization analysis and demonstrates the persistence of local memorization despite existing mitigation methods.

Findings

Memorization extends beyond one-to-one prompt-image pairs.

Existing mitigation methods fail to eliminate local foreground memorization.

A clustering-based mitigation approach shows improved results.

Abstract

Diffusion models (DMs) memorize training images and can reproduce near-duplicates during generation. Current detection methods identify verbatim memorization but fail to capture two critical aspects: quantifying partial memorization occurring in small image regions, and memorization patterns beyond specific prompt-image pairs. To address these limitations, we propose Foreground Background Memorization (FB-Mem), a novel segmentation-based metric that classifies and quantifies memorized regions within generated images. Our method reveals that memorization is more pervasive than previously understood: (1) individual generations from single prompts may be linked to clusters of similar training images, revealing complex memorization patterns that extend beyond one-to-one correspondences; and (2) existing model-level mitigation methods, such as neuron deactivation and pruning, fail to eliminate local memorization, which persists particularly in foreground regions. Our work establishes an effective framework for measuring memorization in diffusion models, demonstrates the inadequacy of current mitigation approaches, and proposes a stronger mitigation method using a clustering approach.