CycleGAN based on tensorflow

Prerequisites

• python 3.5
• tensorflow-gpu 1.4.0
• numpy 1.14.2
• scipy 1.0.0

Datasets

Download the datasets using the following script:

$ ./download_dataset.sh facades

• facades: 400 images from the CMP Facades dataset.
• cityscapes: 2975 images from the Cityscapes training set.
• maps: 1096 training images scraped from Google Maps.
• horse2zebra: 939 horse images and 1177 zebra images downloaded from ImageNet using keywords wild horse and zebra.
• apple2orange: 996 apple images and 1020 orange images downloaded from ImageNet using keywords apple and navel orange.
• summer2winter_yosemite: 1273 summer Yosemite images and 854 winter Yosemite images were downloaded using Flickr API. See more details in our paper.
• monet2photo, vangogh2photo, ukiyoe2photo, cezanne2photo: The art images were downloaded from Wikiart. The real photos are downloaded from Flickr using combination of tags landscape and landscapephotography. The training set size of each class is Monet:1074, Cezanne:584, Van Gogh:401, Ukiyo-e:1433, Photographs:6853.
• iphone2dslr_flower: both classe of images were downlaoded from Flickr. The training set size of each class is iPhone:1813, DSLR:3316. See more details in our paper.

Results

facades X-> Y

RealX	fakeY		fakeY	GenerateX		RealY

apple to orange

apple	G_orange		apple	G_orange		apple	G_orange

Training

$ python3 train.py

If you need help

$ python3 train.py -h

If you want to continue training from the point you halted before, then set 'load_model' parameter.

$ python3 train.py --load_model path_of_your_model

Export model

you can export model for inference

$ python3 export_graph.py --checkpoint_dir path_of_your_checkpoint --XtoY_model facadesX2Y.pb --YtoX_model facedesY2X.pb --image_size 256

Inference

For example:

$ python3 inference.py --model pretrained/your_model.pb --input input.jpg --output output.jpg --image_size 256

Pretrained models

pretrained models of facades: https://github.com/jijeng/CycleGAN/releases

License

This project is licensed under the MIT License - see the LICENSE file for details.

Reference

• The torch implementation of CycleGAN, https://github.com/junyanz/CycleGAN
• the tensorflow implementation of CycleGAN, https://github.com/vanhuyz/CycleGAN-TensorFlow
• original paper : https://arxiv.org/abs/1703.10593