Any advice on how to improve performance on custom datasets?

[Seoung-wook]

The final goal is to increase the recognition rate of Korean and English in Korean, English, Chinese, etc. In the case of the character area, I know that it is recognized in units of sentences. Which do you think will be a better dataset for training, word units or sentence units? Also, is it possible to learn with a Korean-applied dataset and apply it together with an English recognition model?

[yasaslive]

As per documentation and sample dataset you need a word unit for training but I assume you can use sentences units but CRAFT will identify the words so there is no point using sentence.