Cycle GAN in TensorFlow 2.0 with Custom Loops
Implementation of "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks" by Jun-Yan Zhu et al.
generative-adversarial-networks computer-vision image-to-image-translation unpaired-images code notebook paper arxiv:1703.10593 tensorflow tutorial research

  • Readable custom loops to make the implementation consistent with the original paper
  • Tested on Monet2Photo dataset
  • According to the paper, the discriminators were trained using a history of previously generated images rather than ones produced during the current mini-batch. However, to keep it a bit simpler, I did not implement this.
  • Did not decay the learning rate for the second half of the training.

Don't forget to tag @sayakpaul in your comment, otherwise they may not be notified.

Authors
Calling `model.fit()` @ https://pyimagesearch.com | Netflix Nerd
Share this project
Similar projects
Cartoonizer with TensorFlow.js
An app to turn your photos into cartoon-styled images 🎨 within your browsers using White-box Cartoonization GAN.
Breaking the cycle—Colleagues are all you need
A novel approach to performing image-to-image translation between unpaired domains.
Imaginaire
NVIDIA PyTorch GAN library with distributed and mixed precision support.
Paint with Machine Learning
This web app allows you to create a landscape painting in the style of Bob Ross using a deep learning model served using a Spell model server.