Use pretrained model and continue training on own data

[daniiki]

Hi, thanks for this great project. Is it possible to use the pretrained model and continue training on a custom training set containing 5000 images (grey scale images with a dotted font). Do you think the results will be good?

[Bartzi]

Hmm, good question!
I think it depends on some points:

1. are you using the same character set? If yes, it might work quite well
2. is the localization network already able to find the location of single characters? If yes, this is also a good sign.

If none of these two holds. It might be difficult to use a pre-trained model. You would need to have a well performining model with the characters that your dotted fonts contain.
For increasing the number of samples, you could try to generate some artificial samples with the same dotted fonts, if you are able to get hold of such a font.