Chinese Font Creation from Handwriting
Apollys opened this issue · 2 comments
Would it be plausible to use this method to create a font from one's handwriting? If I created a dataset of (2K-3K?) handwritten characters, would you expect it to learn my handwriting style and be able to produce arbitrary characters (my handwriting is pretty neat)? Has anyone tried this? Also, if you have any tips on how to accomplish this (e.g. how to integrate my own images into the workflow), that would be quite helpful!
Edit: after reading some of the recent papers on GANS, I assume this subsequent point is trivial, it should be the default behaviour already. [After learning one's handwriting style, I'm wondering if some part of the model is capable of receiving small amounts of randomness, so that instead of having a font in which every instance of a character is identical, you are able to create many subtly different versions of each characters.]
I think that if these two objectives could be accomplished, the results would be quite fascinating. Furthermore, one could experiment with mixing one's handwriting style with a standard font, or even mixing different people's handwriting styles!
it seems that there is a fork dealing with such situation:
https://github.com/pcgreat/zi2zi
Calling @sungjunyoung and @pcgreat to join in on the issue.