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

2025-05-29

AI Summary

To address detail loss and color distortion in extreme low-light RAW image enhancement, this work pioneers the end-to-end repurposing of a pre-trained generative diffusion model to replace conventional camera ISP pipelines. Methodologically: (1) a physics-aware noise model is formulated directly in the RAW domain; (2) a conditional fine-tuning mechanism enables joint optimization of ISP sub-stages; and (3) leveraging pre-trained weights avoids costly from-scratch training, drastically reducing computational overhead. Compared to existing regression-based or fully trained diffusion approaches, our method achieves new state-of-the-art performance across three major low-light RAW benchmarks—LOL, SID, and FiveK-RAW—demonstrating consistent gains in PSNR, LPIPS, and human perceptual evaluation. Notably, it excels in texture recovery, shadow detail reconstruction, and natural color fidelity.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

High-quality photography in extreme low-light conditions is challenging but impactful for digital cameras. With advanced computing hardware, traditional camera image signal processor (ISP) algorithms are gradually being replaced by efficient deep networks that enhance noisy raw images more intelligently. However, existing regression-based models often minimize pixel errors and result in oversmoothing of low-light photos or deep shadows. Recent work has attempted to address this limitation by training a diffusion model from scratch, yet those models still struggle to recover sharp image details and accurate colors. We introduce a novel framework to enhance low-light raw images by retasking pre-trained generative diffusion models with the camera ISP. Extensive experiments demonstrate that our method outperforms the state-of-the-art in perceptual quality across three challenging low-light raw image benchmarks.

Problem

• Enhancing low-light raw images intelligently
• Overcoming oversmoothing in low-light photos
• Improving sharp details and color accuracy

Innovation

• Retasks pre-trained generative diffusion models
• Enhances low-light raw images
• Integrates with camera ISP