DIF-LIM: A Dual Information Flow Network for Low-Light Image Enhancement.

Low-light image enhancement (LIE) aims to enhance image contrast and recover fine details for images captured in low-light conditions. However, the limitations of using a single image and manually defined priors lead to inadequate information and limited adaptability, resulting in the failure to reveal image details effectively. To tackle this problem, we propose an unsupervised dual information flow LIE model that learns adaptive priors from low-light image pairs. The model is based on the assumption that priors can be learned from pairs of images. Moreover, a simple self+supervised model is designed to perform feature processing on the original input and further processing on the image to avoid the suboptimal limitations of the model. As a result, our model exhibits superior performance on the LIE task compared to other algorithms. As a result, our model exhibits improved network adaptability and superior performance in the LIE task compared to other algorithms.