TS-Diff: Two-Stage Diffusion Model for Low-Light RAW Image Enhancement

2025-05-07

AI Summary

This paper addresses three key challenges in extreme low-light RAW image enhancement: severe noise, color distortion, and poor cross-camera generalization. To this end, we propose TS-Diff, a two-stage diffusion model. In Stage I, multiple virtual cameras are constructed in the noise latent space, and their features are fused via a Camera-aware Feature Integration (CFI) mechanism. In Stage II, a small set of real target-camera images is used to align the model’s noise characteristics with those of the target device. Key innovations include virtual-camera-based noise synthesis, a dynamic global color corrector to suppress chromatic bias, and structural reparameterization for lightweight deployment. We further introduce QID—the first benchmark supporting quantitative illuminance evaluation and wide dynamic range modeling. Extensive experiments on QID, SID, and ELD demonstrate state-of-the-art performance in denoising, cross-camera generalization, and color fidelity, confirming strong potential for practical low-light imaging deployment.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Security and Privacy: Privacy-enhancing technologies
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

This paper presents a novel Two-Stage Diffusion Model (TS-Diff) for enhancing extremely low-light RAW images. In the pre-training stage, TS-Diff synthesizes noisy images by constructing multiple virtual cameras based on a noise space. Camera Feature Integration (CFI) modules are then designed to enable the model to learn generalizable features across diverse virtual cameras. During the aligning stage, CFIs are averaged to create a target-specific CFI$^T$, which is fine-tuned using a small amount of real RAW data to adapt to the noise characteristics of specific cameras. A structural reparameterization technique further simplifies CFI$^T$ for efficient deployment. To address color shifts during the diffusion process, a color corrector is introduced to ensure color consistency by dynamically adjusting global color distributions. Additionally, a novel dataset, QID, is constructed, featuring quantifiable illumination levels and a wide dynamic range, providing a comprehensive benchmark for training and evaluation under extreme low-light conditions. Experimental results demonstrate that TS-Diff achieves state-of-the-art performance on multiple datasets, including QID, SID, and ELD, excelling in denoising, generalization, and color consistency across various cameras and illumination levels. These findings highlight the robustness and versatility of TS-Diff, making it a practical solution for low-light imaging applications. Source codes and models are available at https://github.com/CircccleK/TS-Diff

Problem

• Enhancing extremely low-light RAW images effectively
• Adapting to noise characteristics of specific cameras
• Ensuring color consistency during image enhancement

Innovation

• Two-stage diffusion model for RAW enhancement
• Camera Feature Integration for noise adaptation
• Color corrector ensures dynamic color consistency