SCoRe: Clean Image Generation from Diffusion Models Trained on Noisy Images

TL;DR

SCoRe is a training-free spectral regeneration method that improves image quality from diffusion models trained on noisy data by suppressing high-frequency artifacts and regenerating details at generation time.

Contribution

It introduces a novel spectral cutoff technique combined with SDEdit for clean image generation without retraining, based on a theoretical mapping of cutoff frequency and timestep.

Findings

Outperforms post-processing and noise-robust baselines on CIFAR-10 and SIDD datasets.

Effectively suppresses high-frequency artifacts and restores images closer to clean distributions.

Does not require retraining or fine-tuning of diffusion models.

Abstract

Diffusion models trained on noisy datasets often reproduce high-frequency training artifacts, significantly degrading generation quality. To address this, we propose SCoRe (Spectral Cutoff Regeneration), a training-free, generation-time spectral regeneration method for clean image generation from diffusion models trained on noisy images. Leveraging the spectral bias of diffusion models, which infer high-frequency details from low-frequency cues, SCoRe suppresses corrupted high-frequency components of a generated image via a frequency cutoff and regenerates them via SDEdit. Crucially, we derive a theoretical mapping between the cutoff frequency and the SDEdit initialization timestep based on Radially Averaged Power Spectral Density (RAPSD), which prevents excessive noise injection during regeneration. Experiments on synthetic (CIFAR-10) and real-world (SIDD) noisy datasets demonstrate that SCoRe substantially outperforms post-processing and noise-robust baselines, restoring samples closer to clean image distributions without any retraining or fine-tuning.