DeepVecFont Cyrillic

This is the Pytorch implementation of the paper "Yizhi Wang and Zhouhui Lian. DeepVecFont: Synthesizing High-quality Vector Fonts via Dual-modality Learning. SIGGRAPH Asia. 2021." for Cyrillic letters.

Cyrillic Updates

2023-02-11

• This DeepVecFont modification generates Cyrillic letters with Latin fonts in input.
• Font transformation from TTF/OTF to SFD includes both Latin and Cyrillic lettes.
• Both Main Model and Neural Rasterizer are modified for Cyrillic letters and retrained.
• Rasterizer was trained on 700 epochs with batch size = 12 and learning rate = 0.0001. Main Model was trained on 150 epochs, with batch size = 12 and learning rate = 0.0002.

Future Improvements

• Experiment with Main Model:
  • increasing epoch number;
  • increasing batch size and modifying zero grad in training loop for less powerful GPU;
  • lowering learning rate.
• Considering glyphs with longer sequences (currentlly used max length = 251 is too low for cyrillic leters ж in most fonts)
• Increasing viewport of letters. Currently used viewport crops wide symbols.

Acknowledgment

• DeepVecFont
• SVG-VAE
• SVG-VAE-pytorch by hologerry
• Diffvg
• Paratype