09/05/2022
While working on unpaired style transfer:
Implementation of the Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks paper
by Jun-Yan Zhu, Taesung Park, Phillip Isola, Alexei A. Efros with the help of educational resources.
The model translates apple pictures to the corresponding orange pictures.
Made with Python and Keras.
The model uses:
- a cycle consistency, identity and reconstruction loss
- Instance normalization instead of batch normalization
- a PatchGAN discriminator, which only penalizes structure at the scale of image patches rather than the whole image.
- a U-Net generator (encoder-decoder architecture with skip connections)