PyTorch-CycleGAN

Why we included this project

CycleGAN translates images between two domains without paired examples, turning horses into zebras or photos into paintings, and this repository is a straightforward way to study how that works. The code favors readability over raw speed, so you can trace the two generators and two discriminators and watch how the cycle-consistency loss keeps the round trips consistent. Setup stays light: a script pulls down the standard UC Berkeley datasets, including horse to zebra, apple to orange, and photo to Monet, and the training command starts with the hyperparameters the CycleGAN authors found to work well. You can also point it at your own folder structure if you have a custom style-transfer or domain-adaptation idea in mind. Just know the code targets Python 3.6 and the PyTorch of that era, so treat it as a learning and experimentation base rather than a production pipeline.