SCoRe: Clean Image Generation from Diffusion Models Trained on Noisy Images
2026-04-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors address the problem that diffusion models trained on noisy images often create creepy artifacts in generated images. They propose SCoRe, a method that cleans up these images during generation by cutting out and then carefully regenerating only the noisy high-frequency parts, using a smart way to decide how much to change. Their approach requires no extra training and works well on both fake and real noisy image data, producing cleaner images compared to other techniques. This helps models trained on messy data generate better results without retraining.
diffusion modelsnoisy datasetshigh-frequency artifactsspectral biasfrequency cutoffSDEditRadially Averaged Power Spectral Densityimage generationpost-processingnoise robustness
Authors
Yuta Matsuzaki, Seiichi Uchida, Shumpei Takezaki
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.