Stylized Face Sketch Extraction via Generative Prior with Limited Data (EUROGRAPHICS 2024)

[EG2024] Official repository of StyleSketch [SKSF-A] [Project Page] [Paper]

Getting Started

• install dependency

bash run.sh

• Put styleGAN related checkpoints folder in stylesketch/sketch folder ex) stylesketch/sketch/checkpoints/stylegan_pretrain

  https://drive.google.com/file/d/1X--a491Q6reEBV50XfyYqQ86yDxI44nd/view?usp=drive_link

• Put pretrained StyleSketch weights in model_dir ex) stylesketch/sketch/model_dir https://drive.google.com/file/d/17AgaRzSwXi3c5tmTZztrGGifyHGKrrQu/view?usp=drive_link

How to inference Scripts

Move to sketch folder and run generate.py with the style to extract

cd sketch
python generate.py --train_data sketch_MJ
python generate.py --train_data pencil_sj

How to make the w^+ code to extract sketches?

In our experiment, we used e4e followed by optimization. This can be replaced by different inversion methods.

SKSF-A Sketch Data

SKSF-A consists of seven distinct styles drawn by professional artists, each containing 134 identities and corresponding sketches.

SKSF-A

Acknowledgments

our codes were borrowed from DatasetGAN

If you use this code or SKSF-A for your research, please cite our paper:

@inproceedings{yun2024stylized,
  title={Stylized Face Sketch Extraction via Generative Prior with Limited Data},
  author={Yun, Kwan and Seo, Kwanggyoon and Seo, Chang Wook and Yoon, Soyeon and Kim, Seongcheol and Ji, Soohyun and Ashtari, Amirsaman and Noh, Junyong},
  booktitle={Computer Graphics Forum},
  pages={e15045},
  year={2024},
  organization={Wiley Online Library}
}