CycleGAN on Pytorch

Implementation of CycleGAN on Pytorch. The model learns how to convert an image of domain A to an image of domain B and vice versa. On this project, CelebA dataset has been used as the main dataset. The model has learned how to translate a female image to a male image and vice versa.

Prerequites

• Python 3.6
• PyTorch 0.2.0 (PyTorch is currently available only on Linux and OSX)
• The code has been written on Linux (ubuntu 16.04 LTS) system
• CPU or CUDA-available-GPU

How to

1. Install Python and PyTorch (from the link above or on your own way)

2. Prepare your own dataset, store it in the project folder (it should contain the images of two domains)

3. Change the dataset location part of 'train.py' code like below

...
# The lowest folder should have two sub-folders, each containing images from another domain
image_location = './data/your_dataset/train'
...

4. Run the command below on Terminal

$ python train.py

Results

1) Female to Male

2) Male to Female