RetinexNet

Why we included this project

Low-light photos are a common problem, and this repo is a compact, readable implementation of the RetinexNet method from BMVC 2018. The idea is to split an image into illumination and reflectance layers, brighten the illumination, and keep the detail in the reflectance, which makes it a handy baseline if you are building night-vision or surveillance preprocessing. The model is small enough that training takes minutes on a GPU and stays affordable on CPU, and the included test script lets you point it at a folder of your own images and get enhanced results with minimal setup. Just keep in mind this is research code from 2020, not a maintained product: you will likely need to port the TensorFlow 1.x code before it fits into a modern stack. As a starting point for low-light enhancement, though, it is clear and easy to follow.