AMLS2_CycleGAN

This code is for the final project of ELEL0135_AMLS2. This project works for a style transfer task using CycleGAN, aiming to transform a photo to a painting with style of Monet.

Set up

Environment

Google Colab GPU / Kaggle TPU v3-8

Tensorflow >=2.2

External libraries

tensorflow, tensorflow_addons, matplotlib, IPython, tensorboard

Run the code

Train and generate

Run CycleGAN_Colab.ipynb to train your own model and generate images. Dataset can be downloaded directly by running this code. There are all steps of data preparing, model building, training and generation in this code.

A python version of train and test is also provided as main.py.

Evaluate

FID score is used to evaluate the quality of generated images objectively. Run calculate_fid.ipynb to evaluate your results. Change the path of images to your own dir before you run it.