# RAW-Diffusion: RGB-Guided Diffusion Models for High-Fidelity RAW Image Generation

Source: [RAW-Diffusion: RGB-Guided Diffusion Models for High-Fidelity RAW Image Generation](<https://www.sota2.com/research/paper/68fa75d01e209131cea69fc9>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-28T13:57:11.277Z

Authors: Christoph Reinders, Radu Berdan, Beril Besbinar, Junji Otsuka, Daisuke Iso

Published: 2024-11-20T09:40:12.000Z

## Abstract

Current deep learning approaches in computer vision primarily focus on RGB data sacrificing information. In contrast, RAW images offer richer representation, which is crucial for precise recognition, particularly in challenging conditions like low-light environments. The resultant demand for comprehensive RAW image datasets contrasts with the labor-intensive process of creating specific datasets for individual sensors. To address this, we propose a novel diffusion-based method for generating RAW images guided by RGB images. Our approach integrates an RGB-guidance module for feature extraction from RGB inputs, then incorporates these features into the reverse diffusion process with RGB-guided residual blocks across various resolutions. This approach yields high-fidelity RAW images, enabling the creation of camera-specific RAW datasets. Our RGB2RAW experiments on four DSLR datasets demonstrate state-of-the-art performance. Moreover, RAW-Diffusion demonstrates exceptional data efficiency, achieving remarkable performance with as few as 25 training samples or even fewer. We extend our method to create BDD100K-RAW and Cityscapes-RAW datasets, revealing its effectiveness for object detection in RAW imagery, significantly reducing the amount of required RAW images.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [RGB-to-RAW conversion on MIT-Adobe FiveK Nikon over-exposed (test)](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-mit-adobe-fivek-nikon-over-exposed-test.md>): RGB-to-RAW conversion; MIT-Adobe FiveK Nikon over-exposed (test); PSNR: 32.95.

- [RGB-to-RAW conversion on NOD Sony over-exposed (test)](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-nod-sony-over-exposed-test.md>): RGB-to-RAW conversion; NOD Sony over-exposed (test); PSNR: 35.58.

- [RGB to RAW reconstruction on NOD Nikon (test)](<https://www.sota2.com/research/sota/rgb-to-raw-reconstruction-on-nod-nikon-test.md>): RGB to RAW reconstruction; NOD Nikon (test); PSNR: 39.82.

- [RGB to RAW reconstruction on NOD Sony (test)](<https://www.sota2.com/research/sota/rgb-to-raw-reconstruction-on-nod-sony-test.md>): RGB to RAW reconstruction; NOD Sony (test); PSNR: 38.22.

- [RGB to RAW reconstruction on FIVEK Nikon (test)](<https://www.sota2.com/research/sota/rgb-to-raw-reconstruction-on-fivek-nikon-test.md>): RGB to RAW reconstruction; FIVEK Nikon (test); PSNR: 28.84.

- [RGB to RAW reconstruction on FIVEK Canon (test)](<https://www.sota2.com/research/sota/rgb-to-raw-reconstruction-on-fivek-canon-test.md>): RGB to RAW reconstruction; FIVEK Canon (test); PSNR: 28.89.

- [RGB to RAW reconstruction on PASCALRAW (test)](<https://www.sota2.com/research/sota/rgb-to-raw-reconstruction-on-pascalraw-test.md>): RGB to RAW reconstruction; PASCALRAW (test); PSNR: 35.34.

- [RGB-to-RAW conversion on MIT-Adobe FiveK Canon over-exposed (test)](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-mit-adobe-fivek-canon-over-exposed-test.md>): RGB-to-RAW conversion; MIT-Adobe FiveK Canon over-exposed (test); PSNR: 30.6.

- [RGB-to-RAW conversion on NOD Nikon over-exposed (test)](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-nod-nikon-over-exposed-test.md>): RGB-to-RAW conversion; NOD Nikon over-exposed (test); PSNR: 40.05.

- [RGB-to-RAW conversion on FiveK Canon](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-fivek-canon.md>): RGB-to-RAW conversion; FiveK Canon; PSNR: 40.63.

- [RGB-to-RAW conversion on FiveK Nikon](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-fivek-nikon.md>): RGB-to-RAW conversion; FiveK Nikon; PSNR: 40.35.

- [RGB-to-RAW conversion on NOD Nikon](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-nod-nikon.md>): RGB-to-RAW conversion; NOD Nikon; PSNR: 50.52.

- [RGB-to-RAW conversion on NOD Sony](<https://www.sota2.com/research/sota/rgb-to-raw-conversion-on-nod-sony.md>): RGB-to-RAW conversion; NOD Sony; PSNR: 47.31.

- [sRGB-to-XYZ reconstruction on MIT-Adobe FiveK (test)](<https://www.sota2.com/research/sota/srgb-to-xyz-reconstruction-on-mit-adobe-fivek-test.md>): sRGB-to-XYZ reconstruction; MIT-Adobe FiveK (test); PSNR (Expert A): 19.3.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2411.13150>)

[PDF](<https://arxiv.org/pdf/2411.13150.pdf>)
