Euiyeon-Kim / CycleGAN-Pytorch

Reproduce CycleGAN using pytorch

CycleGAN-Pytorch

Reproduce Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

Main Losses

1. Adversarial Loss

2. Cycle consistency Loss

3. Identity Loss

   • 이게 없어지니까 학습속도가 더뎌짐
   • Paint에서 color composition을 유지하기 위해 사용
   • Input image의 color보존에 도움을 줌