Predict on incomplete glyph set

[matheusrhoden]

Hi!

Is it possible to run the model against an incomplete glyph set?

For example, drawing a few glyphs myself and then having DVF generate the rest of the alphabet.

Thanks!

[yizhiwang96]

Yes, it is exactly the 'few-shot generation' feature demonstrated in our paper. The reference number was 4 in our paper, and theoretically it can be any number, e.g., the number of your drawn glyphs.

[matheusrhoden]

@yizhiwang96 Got it, thanks. Just to clarify, though, do I have to (re)train the model in order to use different reference glyphs? Apparently the modality fusion checkpoint shape has to match the number of reference glyphs