The project implements deep learning models for automatic image colorization, converting images to Lab color space, applying neural network colorization at 256x256 resolution, and upscaling back to full resolution. It includes both the original ECCV 2016 model and the interactive SIGGRAPH 2017 extension. The repo now offers a PyTorch implementation alongside the historical Caffe code.