yizhiwang96 / deepvecfont

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

Question abount some characters (e.g., g, p, q ...), The parts below these characters baselines are cut #33

Open graphl opened 1 year ago

graphl commented 1 year ago

imageimageimage The 3 SVGs generated above are g, j, and q,None of the parts below their baseline are generated.

Question :

Is there any way to make these words display normally? If I want to get the generated font baseline information, do I need to convert the generated SVG and BMP to fonts?

yizhiwang96 commented 1 year ago

This is an issue brought by diffvg refinement. In DeepVecFont-v2, there will be no such issue. The code is expected to be released in one week, please stay tuned.

graphl commented 1 year ago

Thank you very much for your reply, I'm looking forward to the DeepVecFont-v2 to be released? I'd like to ask when will the code for be released?

@yizhiwang96

yizhiwang96 commented 1 year ago

Thank you very much for your reply, I'm looking forward to the DeepVecFont-v2 to be released? I'd like to ask when will the code for be released?

@yizhiwang96

Code has been released.

lqb17909 commented 1 year ago

You said This is an issue brought by diffvg refinement. Can you explain it? I would like to understand why diffvg refinement can cause this issue。 @yizhiwang96

yizhiwang96 commented 1 year ago

You said This is an issue brought by diffvg refinement. Can you explain it? I would like to understand why diffvg refinement can cause this issue。 @yizhiwang96

To be accurate, DiffVG is not the main reason. First, I followed the rendered style of SVG-VAE (letter baseline is close to the bottom of images), causing letters like p and q have some parts outside the image canvas. Then I used DiffVG to align SVGs with the images, where SVGs are overfitted to the incomplete images, causing this problem. To solve this issue, you could try to render images with a size of 64 128 (width height), where p and q can be shown in images with complete shapes.