search

yizhiwang96 / deepvecfont

[SIGGRAPH Asia 2021] DeepVecFont: Synthesizing High-quality Vector Fonts via Dual-modality Learning
MIT License
182 stars 31 forks source link

Questions about the glyph images #5

Closed hyh-dev closed 2 years ago

hyh-dev commented 2 years ago

Hi, I found that some glyph images in ground truth are missing some of the handwriting. Where did i go wrong ？ image

yizhiwang96 commented 2 years ago

Hi, Yuhao. It is not wrong. We fix the glyph baseline of each font. For some characters (e.g., g, p, q ...), the lower parts may be missing in a square canvas (64 64). It does not affect the generation of vector glyphs, but could affect the refinement process. A solution is that render longer images (such as 64 128) or lift the baseline of these characters.

graphl commented 1 year ago

@yizhiwang96 Why do you need to fix baselines? What happens if I keep the data below the baseline? If I want to get the generated font baseline information, do I need to convert the generated SVG and BMP to fonts?