/chainer-cyclegan

Unofficial chainer implementation of 'Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks' [Zhu+, ICCV2017]

Primary LanguagePython

Chainer-CycleGAN

This is an unofficial chainer re-implementation of a paper, Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks. This implementation is based on this

Requirements

  • Python 3.5+
  • Chainer 2.0+
  • ChainerCV 0.7+
  • Numpy
  • Matplotlib

Usage

Download and expand dataset

Downloadable datasets are listed in ./datasets/download_cyclegan_dataset.sh

./datasets/download_cyclegan_dataset.sh <dataset>

Training

python train.py --load_dataset <dataset> --gpu <gpu>

Test for single image

python single_image_test.py <input_image> --gpu <gpu> --load_gen_model <trained_generator> --output <output_image>

left: input (horse), right: output (zebra)

input(horse) output (zebra)

References