CycleGAN-pytorch

code based on junyanz/pytorch-CycleGAN-and-pix2pix

TODO

• efficient-det as discriminator backbone
• trivial upsample generator

Command Line

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.

If you run without it is fine. The code will run the command automaitc

Train the model

python train.py --dataroot ./dataset/selfie2anime --name selfie2anime_ustyle --model ustyle

python train.py --dataroot ./dataset/selfie2anime --name selfie2anime_ustyle --model ugatit --light

Continue training

python train.py --dataroot ./dataset/selfie2anime --name selfie2anime_ustyle --model ustyle --continue_train

python train.py --data_root ./horse2zebra --name h2z_cyclegan --model cycle_gan

Test the model

python test.py --dataroot ./dataset/selfie2anime --name selfie2anime_ustyle --model ustyle

python test.py --data_root ./horse2zebra --name h2z_cyclegan --model cycle_gan