Add an option to choose trained language model types (fast,normal,best)

[vivadavid]

Describe your problem:

Hi,

Given the space taken by the language packages, I presume Normcap downloads the fast Tesseract trained models. I wonder if it would be a good idea to download, instead, the standard trained models. They take only a few more megabytes and they should be more precise, but I haven't tested anything and I don't know if they take too many resources. If we were talking about performing OCR on all the pages of a book, using the fast model on a slow computer might save some time, but when it comes to a screenshot, the impact might be negligible.

Solution you'd like to see:

No response

Alternatives you considered:

No response

Additional information or remarks:

No response

[dynobo]

Hi @vivadavid , thanks for your suggestion! :-)

You are correct, currently the "fast" models are used by NormCap by default. The reason is not file size but detection speed. They are significantly faster, with only minor detection quality degradation. Afaik, the language models provided as packages by common Linux distribution also ship the "fast" models.

Also worth mentioning: As a NormCap user, you already can use the "normal" models as well as the even slower "best" models, but you have to download them manually (see here, point 6.))

To conclude: I think "fast" models should stay the default. But I'm very open for contributions to add an option to the Language Manager to choose between the trained model types fast/normal/best!

[vivadavid]

I understand your points, and it's great that I could manually replace the trained models if necessary.