DarkDiff: Advancing Low-Light Raw Enhancement by Retasking Diffusion Models for Camera ISP

Amber Yijia Zheng, Yu Zhang, Jun Hu, Raymond A. Yeh, Chen Chen · 2025

Read DarkDiff: Advancing Low-Light Raw Enhancement by Retasking Diffusion Models for Camera ISP by Amber Yijia Zheng, Yu Zhang, Jun Hu, Raymond A. Yeh, Chen Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-23743.txt.

Description

Amber Yijia Zheng and co-authors present a 2025 method, DarkDiff, for rescuing photos taken in near darkness. Camera software that cleans raw dark images often smooths detail. Models trained from scratch for the task can miss sharp detail and true color. DarkDiff instead retasks a ready-made image generation model for the camera pipeline. The pre-trained model's sense of what scenes look like guides the clean-up. Noisy raw input goes in, and a cleaner, sharper image comes out.

Tests on three low-light data sets led the field in judged quality. The method kept fine detail that regression-based tools tended to blur. Colors in deep shadow came out closer to the truth. The authors say reuse of a strong general model beats building a narrow one. That lesson may shape how camera software is built next. The work targets the raw stage, before photo processing.

Catalog record

CREATORAmber Yijia Zheng, Yu Zhang, Jun Hu, Raymond A. Yeh, Chen Chen
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-23743

Browse the full library · About the Tower of Babel