CA-GAN

We provide PyTorch implementation for our CYB paper "Towards Realistic Face Photo-Sketch Synthesis via Composition-Aided GANs"

[paper]

Our Proposed Framework

Sample Result

left: sketch synthesis; right: photo synthesis

(a)Input Image, (b)cGAN, (c)CA-GAN, (d)SCA-GAN

Citation

If you find this useful for your research, please use the following.

@article{gao2017ca-gan,
	title = {Towards Realistic Face Photo-Sketch Synthesis via Composition-Aided GANs},
	author = {Jun Yu, Xingxin Xu, Fei Gao, Shengjie Shi, Meng Wang, Dacheng Tao, and Qingming Huang},
	booktitle = {arXiv:1712.00899},
	year = {2019},
}

Prerequisites

• Linux or similar environment
• Python 2.7
• NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Clone this repo:

  git clone https://github.com/jehovahxu/ca-gan
  cd ca-gan

• Install PyTorch 0.4+ and torchvision from http://pytorch.org and other dependencies (e.g., visdom and dominate). You can install all the dependencies by

  pip install -r requirments.txt

ca-gan train/test

• Download a dataset(CUFS split train and test with this files)

• Download the VGG-Face model which based on the VGG-Very-Deep-16 CNN architecture. Here we convert torch weight to pyTorch to fit our frame, you can download our converted model directly.

  • Google Drive

• Get face parsing

  • here we use Face Labling to get face parsing
  • Check out the branch to get the our newly used

• Train a model

  python main.py --model_vgg {model path}

• Test the model

  python test.py --dataroot {data path} --fold {epoch number}

  • The option fold is used for load ./checkpoint/netG_epoch_'+fold+'.weight and you can edit it in test.py

Apply a pre-trained model

• You can download a pre-trained model which trained on the CUSF and can transform photo to sketch :

  • Google Drive

• The pre-trained model need to be save at ./checkpoint and named it as netG_epoch_'+fold+'.weight

• Then you can test the model

Datasets

• CUFS
• CUFSF

Result

• Our final result can be downloaded:
  • Google Drive

Training/Test Tips

Best practice for training and testing your models.

Feel free to ask any questions.Xingxin Xu, jehovahxu@gmail.com

Acknowledgments

Our code is inspired by pytorch-CycleGAN-and-pix2pix