Progressive training of StarGAN along with consistency loss term
This project is an implementation of clubbing 3 different papers namely, StarGAN, Progressive Training of GANs and CT-GAN. Each paper provides a contribution to achieve high resolution photo-realism.
This code has been built over the original code provided by the authors of StarGAN. This GAN is trained for the same task as that of StarGAN i.e. image translation. The datasets I used are CelebA, CelebA-HQ that I obtained from this script and RafD.
The results I obtained after training it for 38 hours on a Titan-V -