Training on custom dataset

[filipaldi]

Hi @yizhiwang96,

I am doing a research project where I want to target sequential irregularities with a specific dataset. I'd like to test the hypothesis on the DeepVecFont-2 system.

I have a question regarding the dataset preparation.

1. Do I have to convert .otf files to .sdf files using your script, or can I use my scripts? If yes, could you explain the reasons?
2. As first I've prepared .sdf files, can I use them directly instead of exporting to .otf/.ttf and then again converting to .sdf?

My dataset consists of composite glyphs, which are assemblages of other glyphs or components. Exporting to .otf/.ttf causes loss of such data.
More info about the composite glyphs here: https://fontforge.org/docs/techref/accented.html