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

Yuta Matsuzaki; Seiichi Uchida; Shumpei Takezaki

cs.CVApr 10, 2026

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

Authors: Yuta Matsuzaki, Seiichi Uchida, Shumpei Takezaki

Organizations: Kyushu University, Fukuoka, Japan

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.

Figures & tables

Fig. 1: Impact of a dataset containing noisy training images on the generated results. Images in red boxes are clean examples; all others are noisy.

Fig. 2: (a) Overview of the diffusion and reverse processes in terms of frequency components. (b) SCoRe : training-free, generation-time spectral regeneration that suppresses high-frequency components via a frequency cutoff and regenerates high-frequency details via SDEdit.

Fig. 3: The generation process of a diffusion model trained on noisy images. (a) Standard generation process: produces noisy generated images. (b) SDEdit: When SDEdit is applied to a noisy generated image, the result is regenerated as a noisy image. (c) SCoRe: Removal of high-frequency components via a cutoff operation, followed by regeneration using SDEdit to yield cleaner images.

Fig. 4: Generated images under CIFAR-10 with Gaussian noise: (a) Training examples, (b) standard diffusion sampling, (c) generated images post-processed with a bilateral filter, (d) generated images post-processed with Noise2Void, (e) NR-GAN, and (f) SCoRe.

Fig. 5: Comparison of the RAPSDs of clean images, training images, and the results produced by the proposed method under Gaussian noise.

Method	Gaussian	Poisson	Mix
Diffusion Model	132.9	168.7	203.2
w/ Post-Bilateral	51.7	68.0	102.8
w/ Post-N2V	32.3	31.6	40.1
SDEdit	132.2	151.3	180.2
NR-GAN	15.5	46.5	32.0
SCoRe ( $f_{\mathrm{cutoff}}=0.20$ , $t^{\prime}=119$ )	14.0	14.9	16.2

TABLE I: Comparison of FID on synthetic noisy dataset

Fig. 6: Effect of the noisy-data ratio in training on FID (x-axis: percentage of noisy training images).

Fig. 7: Generated Results : (a) training examples, (b) standard diffusion sampling, (c) generated images post-processed with a bilateral filter, (d) generated images post-processed with Noise2Void, (e) NR-GAN, and (f) SCoRe.

Method	SIDD
Diffusion Model	30.1
w/ Post-Bilateral	30.9
w/ Post-N2V	27.0
SDEdit	30.6
NR-GAN	46.3
SCoRe ( $f_{\mathrm{cutoff}}=0.05$ , $t^{\prime}=110$ )	29.0

TABLE II: Comparison of FID on real-world noisy dataset

Diffusion-Based Image Generation

Diffusion-based image generation creates images by progressively denoising noise, with controls for content, style, and structure.

See all 324 papers

May 18, 2026

Spectral Progressive Diffusion for Efficient Image and Video Generation

Diffusion models have been shown to implicitly generate visual content autoregressively in the frequency domain, where low-frequency components are generated earlier in the denoising process while high-frequency details emerge only in later timesteps. This structure offers a natural opportunity for efficient generation, as high-resolution computation on noise-dominated frequencies is largely redundant. We propose Spectral Progressive Diffusion, a general framework that progressively grows resolution along the denoising trajectory of pretrained diffusion models. To this end, we develop a spectral noise expansion mechanism and derive an optimal resolution schedule from the model's power spectrum. Our framework supports training-free acceleration and a novel fine-tuning recipe that further improves efficiency and quality. We demonstrate significant speedups on state-of-the-art pretrained image and video generation models while preserving visual quality.

Howard Xiao, Brian Chao, Lior Yariv +1

Stanford University

Video Diffusion Models Diffusion Sampling Diffusion-Based Image Generation Diffusion Model Inference Acceleration Diffusion Models

Ask PDF

Mar 19, 2026

Spectrally-Guided Diffusion Noise Schedules

Denoising diffusion models are widely used for high-quality image and video generation. Their performance depends on noise schedules, which define the distribution of noise levels applied during training and the sequence of noise levels traversed during sampling. Noise schedules are typically handcrafted and require manual tuning across different resolutions. In this work, we propose a principled way to design per-instance noise schedules for pixel diffusion, based on the image's spectral properties. By deriving theoretical bounds on the efficacy of minimum and maximum noise levels, we design ``tight'' noise schedules that eliminate redundant steps. During inference, we propose to conditionally sample such noise schedules. Experiments show that our noise schedules improve generative quality of single-stage pixel diffusion models, particularly in the low-step regime.

Carlos Esteves, Ameesh Makadia

Google Research.

Diffusion-Based Image Generation Diffusion Timestep Scheduling Diffusion-Based Denoising Pixel-Space Diffusion Models Diffusion Model Inference Acceleration

Ask PDF

Mar 2, 2026

Spectral Regularization for Diffusion Models

Diffusion models are typically trained using pointwise reconstruction objectives that are agnostic to the spectral and multi-scale structure of natural signals. We propose a loss-level spectral regularization framework that augments standard diffusion training with differentiable Fourier- and wavelet-domain losses, without modifying the diffusion process, model architecture, or sampling procedure. The proposed regularizers act as soft inductive biases that encourage appropriate frequency balance and coherent multi-scale structure in generated samples. Our approach is compatible with DDPM, DDIM, and EDM formulations and introduces negligible computational overhead. Experiments on image and audio generation demonstrate consistent improvements in sample quality, with the largest gains observed on higher-resolution, unconditional datasets where fine-scale structure is most challenging to model.

Satish Chandran, Nicolas Roque dos Santos, Yunshu Wu +2

Department of Mathematics, University of California Riverside, Riverside, California, USA · Department of Computer Science, University of California Riverside, Riverside, California, USA

Diffusion Models Spectral Regularization Diffusion-Based Image Generation Audio Generation

Ask PDF

Abstract

Diffusion-Based Image Generation

Explore similar work

Spectral Progressive Diffusion for Efficient Image and Video Generation

Spectrally-Guided Diffusion Noise Schedules

Spectral Regularization for Diffusion Models