Efficient HDR Reconstruction From Real-World Raw Images

High dynamic range (HDR) imaging is a significant yet challenging problem due to the limited dynamic range of generic image sensors. Most existing learning-based HDR reconstruction methods take a set of bracketed exposure sRGB images to extend the dynamic range. However, they overlook the computational and memory inefficiencies of Image Signal Processors (ISPs) when processing a set of sRGB images with different exposures. Furthermore, the absence of large-scale raw-based HDR datasets limits the research on HDR imaging. In this work, in a new aspect, we discover an excellent opportunity for HDR reconstructing directly from raw images and investigating novel neural network structures that benefit the deployment of mobile devices. Meanwhile, we construct a new HDR dataset containing raw images and process to obtain sRGB images and design a new model to reconstruct HDR utilizing the unique characteristics of long- and short-exposure images. Our key insights are threefold: (1) a new computational raw LDR-HDR pair formation pipeline is designed to construct a real-world raw HDR dataset called RealRaw-HDR; (2) a lightweight-efficient HDR model, RepUNet, is developed using the structural reparameterization technique; (3) a plug-and-play alignment-free and motion-aware short-exposure-first selection loss and a colorfulness loss are proposed to mitigate ghost artifacts and color cast. Extensive experiment results demonstrate that our approach achieves state-of-the-art performance in both visual quality and quantitative metrics