CycleGAN in PyTorch

I use the cyclegan code provided by https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix .It allows users who have never applied makeup before to easily try other people's makeup without spending time learning makeup techniques and money on cosmetics, and then make a reference for their own makeup style.

BTW:I use the Bilinear_interportation+Reflection2d+Conv2d instead of Con2dTranspose in the generator to prevent the checkerboard problem.

Prerequisites

• Linux or macOS
• Python 3
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

pip install -r requirements.txt

CycleGAN train/test

• To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097.

• Train a model:

python train.py --dataroot ./Train_Test_datasets/makeup --name makeup_cyclegan --model cycle_gan

To see more intermediate results, check out ./checkpoints/makeup_cyclegan/web/index.html.

• Test the model:

python test.py --dataroot ./Train_Test_datasets/makeup --name makeup_cyclegan --model cycle_gan

• The test results will be saved to a html file here: ./results/makeup_cyclegan/latest_test/index.html.