Closed forzagreen closed 1 year ago
What exactly are you looking for?
For the training with artificial data, there already is a Python package (https://github.com/tesseract-ocr/tesstrain/tree/main/src, tesstrain
on PyPI with some smaller modifications, currently maintained/owned by me in a fork of the original code).
For the training with real data, there currently mostly is a Makefile. If I remember the discussions in some PRs correctly, one collaborator has some plans about moving everything to Python and providing it in one package, but there are no results for this at the moment.
That being said, I see no real value in pytesseract adding functionality like this.
Hi @stefan6419846 , thank you for sharing these information. I didn't know about this pypi package and the python code behind it.
The documentation of training is confusing and scattered between 3 repos (tesseract, tessdoc and tesstrain). It documentes only Makefiles. It's worth documenting the python options.
Thanks again. Closing this issue.
tessdoc documents the training process with the Python package in a basic manner without any actual references to tesstrain
(or the tesstrain.sh
script, which was the old way): https://tesseract-ocr.github.io/tessdoc/tess5/TrainingTesseract-5.html But yes, I already mentioned the not very clear docs in the past, but priority does not seem to be high for it and my experience is rather restricted to the training with artificial data.
It would be great if pytesseract offers a wrapper around the training functionalities of Tesseract (https://github.com/tesseract-ocr/tesstrain) Since the training is not done often in Tesseract, the option can be added as a package extras, e.g. installed as
pip install pytesseract[training]