Tesseract training - font size

[Honza]

Hi all,

Is the font size important for a training of a new model from scratch? I found that it is not important for the Tesseract v3.0 but I was not able to find anything about Tesseract v.4 and later.

Right now I use default parameters:

--xsize 3600 for tesstrain.sh script

--ptsize 12 for text2image in tesstrain.sh

The net spec is defined like this: 

'[1,0,0,3Ct3,3,16Mp3,3Lfys64Lfx96Lrx96Lfx192O1c14]' (variable input height and width)

Do I need to train the model on multiple font sizes (i.e. by finetunning)?

Thanks.

Jan