Korean YetHangul(medival korean) Font Generator Using StarGAN v2

used code below

StarGAN v2 — Official TensorFlow Implementation [Paper] [Pytorch]

Implemented by Junho Kim

CAUTION

• This project is working on very specific case only.
• Font generating was only tested on 64*64 RGB image
• Some korean font are needed to run properly
• Before training, you must generate dataset

Requirements

• Tensorflow == 2.1.0
• Tensorflow-addons == 0.9.1
• opencv-python
• Pillow
• tqdm
• gpuutil
• humanize
• fonttools
• 네이버 나눔명조옛한글 (https://hangeul.naver.com/2014/archaicword)
• 네이버 나눔바른고딕옛한글 (https://hangeul.naver.com/2014/archaicword) optional
• 배달의민족 을지로체 (https://www.woowahan.com/) optional
• 네이버 나눔고딕 (https://hangeul.naver.com/font) optional
• 네이버 나눔고딕 (https://hangeul.naver.com/font) optional

Usage

├── dataset
   └── YOUR_DATASET_NAME
       ├── train
           ├── domain1 (domain folder)
               ├── xxx.jpg (domain1 image)
               ├── yyy.png
               ├── ...
           ├── domain2
               ├── aaa.jpg (domain2 image)
               ├── bbb.png
               ├── ...
           ├── ...
           
       ├── test
           ├── ref_imgs (domain folder)
               ├── domain1 (domain folder)
                   ├── ttt.jpg (domain1 image)
                   ├── aaa.png
                   ├── ...
               ├── domain2
                   ├── kkk.jpg (domain2 image)
                   ├── iii.png
                   ├── ...
               ├── ...
               
           ├── src_imgs
               ├── src1.jpg 
               ├── src2.png
               ├── ...

        ├── practice
           ├── ref_imgs (domain folder)
               ├── domain1 (domain folder <<-- target font style)
                   ├── ttt.jpg (domain1 image <<-- 1 image. more than 1 is meaningless)

           ├── src_imgs (<<-- will be automatically generated)
               ├── text.txt (characters you want to generate)

Train

python main.py --dataset yet_hangul --phase train --img_size 64

Test

python main.py --dataset yet_hangul --phase test --img_size 64

Practice

When you write yet hangul, each character must be one unicode. If not, program might read as 2 or more characters, rather than 1 character. Be careful if you copy from internet. Many yet hangul words written on internet are combinational. (ᄎᆞᆷ〮 <-- .+ㅊ+ . +ㅁ) Use this webpage would be helpful (https://www.korean.go.kr/common/oldHangeul.do)

python main.py --dataset yet_hangul --phase practice --img_size 64

GenerateDataset.ipynb

click RUN ALL button...

Tensorflow results (20K)

Latent-guided synthesis

yet_hangul

Reference-guided synthesis

yet_hangul

Pretrained checkpoint

https://drive.google.com/file/d/1dXIQG5baItHa6-5W9ZNZnJi_1OK8lol8/view?usp=sharing

License

The source code, pre-trained models, and dataset are available under Creative Commons BY-NC 4.0 license by NAVER Corporation. You can use, copy, tranform and build upon the material for non-commercial purposes as long as you give appropriate credit by citing our paper, and indicate if changes were made.

For business inquiries, please contact clova-jobs@navercorp.com.
For technical and other inquires, please contact yunjey.choi@navercorp.com.
For questions about the tensorflow implementation, please contact jhkim.ai@navercorp.com.

Citation

If you find this work useful for your research, please cite our paper:

@inproceedings{choi2020starganv2,
  title={StarGAN v2: Diverse Image Synthesis for Multiple Domains},
  author={Yunjey Choi and Youngjung Uh and Jaejun Yoo and Jung-Woo Ha},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2020}
}