This repository features several notebooks exploring the performance of different GAN architectures in the task of translating an outline image of a face into a realistic 256x256 image of a human face.
models/conditional_gan contains the code used to fit a conditional GAN to our data.
models/cycle_gan contains the code used to fit a cycle consistent GAN to our data.
models/final_gan contains the code for our proposed model architecture, which combines a conditional GAN approach with deep learning based upscaling.
Facial images taken from https://www.kaggle.com/jessicali9530/celeba-dataset.
Realistic sketches generated using openCV.
