CycleGAN for Image Translation

This repository aims at developing a CycleGAN model to translate photos of horses to zebras and vice versa. Here, it is not necessary to provide paired examples in the dataset to CycleGAN, i.e. it does not require examples of photographs before and after the translation in order to train the model. Instead, the model is able to use a collection of photographs from each domain and extract and harness the underlying style of images in the collection in order to perform the translation.

Requirements

numpy
matplotlib
keras

Usage

Data

The horse2zebra dataset can be downloaded from the CycleGAN webpage.
In the dataset, the A category refers to horse and B category refers to zebra, and the dataset is comprised of train and test elements. However, the current implementation loads all photographs and use them as a training dataset.
All photographs are of the same size, i.e. 256 * 256.
To prepare the image dataset, considered here, in order to properly train the CycleGAN model, run Data.py

Training and Testing

To train and test the CycleGAN model along with plotting the results, run Cycle_GAN_to_translate_Images.py
To translate multiple images using ad-hoc image translation process with the help of trained CycleGAN, run Image_Translation_with_Cycle_GAN_Generators.py
To translate merely individual images using the trained CycleGAN model, run Single_Image_Translation_with_Cycle_GAN_Generators.py
All hyperparameters to control training and testing of the CycleGAN model in single as well as multiple image translation settings are provided in their respective .py files.
The average loss for both generator and discriminator components of the CycleGAN model for each of the two catagories (Horse (A) and Zebra (B)) are printed after every epoch.