In this repository, we provide Generative Models to adress the Image Colorization problem.
The generative network is given a high resolution 512x512 input image in the B&W domain and generates the corresponding 512x512 image in the RGB domain.
We followed two approaches, the first one is in a supervised way where we trained a colorization GAN, where the loss is computed by comparing the output of the generative network with the ground truth.
The other approache is unsupervised, where we trained a cycle GAN as proposed in the original paper https://arxiv.org/pdf/1703.10593.pdf but we omitted the use of the identity loss as the images of the two generators don't have the same output channels.
P.S.1: feel free to comment and contribute to this modest work ;-)
P.S.2: pretrained models will be added soon (-: