SpiralDiff: Spiral Diffusion with LoRA for RGB-to-RAW Conversion Across Cameras

TL;DR

SpiralDiff is a diffusion-based framework with camera-aware adaptation for high-quality RGB-to-RAW conversion, addressing intensity-dependent reconstruction challenges and multi-camera variability, improving downstream detection tasks.

Contribution

The paper introduces SpiralDiff, a novel diffusion model with signal-dependent noise weighting and CamLoRA for camera-specific adaptation in RGB-to-RAW conversion.

Findings

Outperforms existing methods in conversion quality

Enhances RAW-based object detection accuracy

Demonstrates effectiveness across multiple datasets

Abstract

RAW images preserve superior fidelity and rich scene information compared to RGB, making them essential for tasks in challenging imaging conditions. To alleviate the high cost of data collection, recent RGB-to-RAW conversion methods aim to synthesize RAW images from RGB. However, they overlook two key challenges: (i) the reconstruction difficulty varies with pixel intensity, and (ii) multi-camera conversion requires camera-specific adaptation. To address these issues, we propose SpiralDiff, a diffusion-based framework tailored for RGB-to-RAW conversion with a signal-dependent noise weighting strategy that adapts reconstruction fidelity across intensity levels. In addition, we introduce CamLoRA, a camera-aware lightweight adaptation module that enables a unified model to adapt to different camera-specific ISP characteristics. Extensive experiments on four benchmark datasets demonstrate the superiority of SpiralDiff in RGB-to-RAW conversion quality and its downstream benefits in RAW-based object detection. Our code and model are available at https://github.com/Chuancy-TJU/SpiralDiff.