• Prerequisites
• Getting Started
• • Installation
  • SDGAN train/test
  • Model
  • Results show
• Citation

Prerequisites

• Python 3
• CPU or NVIDIA GPU + CUDA

Getting Started

Installation

• Clone this repo:

git clone https://github.com/qinghew/SDGAN
cd SDGAN
pip install -r requirements.txt

SDGAN train/test

• Train a model:

python train.py --dataroot dataset_root --name SDGAN

• The training results and loss plots are saved to here: ./results/SDGAN. cd there, run tensorboard --logdir=./ --port=6006 and click the URL http://localhost:6006.

• Test the model:

#!./scripts/test_cyclegan.sh
python test.py --dataroot dataset_root --name SDGAN --gpu_ids 0 --aspect_ratio 0 --model cycle_gan

• The test results will be saved to here: ./results/SDGAN/test_latest/.

Model

• SDGAN:
  

• Omni-directional pixel-gradient convolution kernel:
  

Results show

• Results on photo→sketch:
  

• Results on sketch→photo:
  

• Results on APDrawing⇔photo:
  

• Results on wild data:
  

• Results on the constructed unpaired datasets:
  

Citation

If you use this code for your research, please cite our papers.

@article{wang2022SDGAN,
  title={Self-Discriminative Cycle Generative Adversarial Networks for Face Image Translation},
  author={Wang, Qinghe and Cao, Bing and Zhu, Pengfei and Wang, Nannan and Hu, Qinghua and Gao, Xinbo},
  journal={SCIENTIA SINICA Informationis},
  pages={DOI: 10.1360/SSI-2021-0321}, 
  year={2022}
}