YeTI: You Only Need Two Noisy Images for Real-World sRGB Noise Generation

2026-07-10

AI Summary

Real-world sRGB image denoising is highly challenging due to the non-linear nature of sensor noise and the absence of paired clean–noisy image data. This work proposes a method that requires only two unpaired noisy images of the same scene, leveraging a reconstruction autoencoder to disentangle content from noise and a one-step conditional diffusion Transformer to model signal-dependent noise distributions. For the first time, it enables realistic sRGB noise synthesis without clean references or camera metadata. A consistency-based training objective is introduced to significantly enhance noise modeling accuracy. The realism of the generated noise is validated on real-world datasets including SIDD, SIDD+, MAI2021, and SID, and denoising models trained on YeTI-synthesized data achieve state-of-the-art performance on the DND benchmark.

Abstract

Real-world sRGB image denoising remains challenging due to the nonlinear characteristics of sensor noise and the difficulty of acquiring aligned clean-noisy image pairs. Supervised denoisers often overfit to limited paired datasets, while self-supervised methods still depend on sufficiently diverse noisy observations. These limitations motivate scalable noise synthesis methods that can model real-world noise without clean ground truth or camera metadata. We propose YeTI, a real-world sRGB noise generation framework that learns from only two noisy observations of the same scene. YeTI uses a Reconstruction Autoencoder to disentangle scene structure and noise characteristics, and models the latent noise distribution with a one-step Conditional Diffusion Transformer trained using consistency objectives. Given a single noisy input at inference time, YeTI generates realistic, signal-dependent noise while preserving the underlying scene content. Extensive experiments demonstrate the effectiveness of YeTI across real-world benchmarks. We evaluate noise generation on SIDD and further assess generalization on SIDD+, MAI2021, and SID, covering smartphone and diverse consumer-camera sensors. Downstream denoising results on DND further show that denoisers trained with YeTI-synthesized images achieve strong real-world performance, highlighting the practical value of clean-image-free and metadata-free noise generation.

Problem

• real-world sRGB denoising
• noise synthesis
• clean-noisy pairs
• sensor noise
• metadata-free

Innovation

• noise generation
• two-noisy-image learning
• conditional diffusion transformer
• self-supervised denoising
• real-world sRGB noise