pGAN icon indicating copy to clipboard operation
pGAN copied to clipboard

The code of paper: Robust Face Sketch Synthesis via Generative Adversarial Fusion of Priors and Parametric Sigmoid (pGAN) [IJCAI 2018]

trafficstars

This is the project page associating to our work on Robust Face Sketch Synthesis:

Zhang, S., Ji, R., Hu, J., Gao, Y., Lin, C. W., "Robust Face Sketch Synthesis via Generative Adversarial Fusion of Priors and Parametric Sigmoid." In IJCAI, 2018.

This page contains the codes for our model "pGAN". If you have any problem, please feel free to contact us.

Prerequisites

  • Python (2.7 or later)
  • numpy
  • scipy
  • NVIDIA GPU + CUDA 8.0 + CuDNN v5.1
  • pyTorch 0.3

Getting Priors

  • PortraitFCN [1]
  • P-Net [2]

Training & Test

After preparing the priors and training/test images, run:

./run.sh

The example of runing the training phase:

python train.py --dataset CUHKStudent --nEpochs 20 --cuda

The example of runing the test phase:

python test.py --train_data CUHKStudent --test_data XM2VTS --G1_model G_1_model_epoch_20.pth --G2_model G_2_model_epoch_20.pth --my_layer_model my_layer_model_epoch_20.pth --cuda

Pretrained Model & Preprocessed Data

The pretrained model and preprocessed data can be found at: Google Drive or Baidu Netdisk (password: gen3).

Reference

[1] Shen, X., Hertzmann, A., Jia, J., Paris, S., Price, B., Shechtman, E., Sachs, I., "Automatic portrait segmentation for image stylization." In Computer Graphics Forum, 2016.

[2] Liu, S., Yang, J., Huang, C., Yang, M. H., "Multi-objective convolutional learning for face labeling." In CVPR, 2015.

[3] Zhang, S., Ji, R., Hu, J., Gao, Y., Lin, C. W., "Robust Face Sketch Synthesis via Generative Adversarial Fusion of Priors and Parametric Sigmoid." In IJCAI, 2018.

Citation

If our paper helps your research, please cite it in your publications:

@inproceedings{zhang2018robust,
  title={Robust Face Sketch Synthesis via Generative Adversarial Fusion of Priors and Parametric Sigmoid.},
  author={Zhang, Shengchuan and Ji, Rongrong and Hu, Jie and Gao, Yue and Lin, Chia-Wen},
  booktitle={IJCAI},
  pages={1163--1169},
  year={2018}
}