Low light image enhancement using Deep Retinex-Net model

The problem is to train a Deep Retinex-Net model on the dataset containing the pair of low and high light images and make the model be able to convert any given low light image into a high light image. The dataset named LOL (LOw Light paired) dataset used for the problem is taken from the above link. It contains 5000 low/normal light images pairs of different kinds such as household appliances, toys, books, garden, food items, playground, clubs, streets, etc. These raw images are resized to 128*128 and converted to Portable Network Graphics format. The below figure shows the subset of these images.