The Best way to fine tune a Tesseract model

[Omar Samir]

Now I am trying to fine tune Tesseract Arabic model in the testdata_best with more than 20K images in the dataset. I want to know the best values for parameters like MAX_ITERATIONS, EPOCHS, and LEARNING_RATE so how can I find these values I also look inside research papers but I find nothing about the best values for training a model.

[La Monte H. P. Yarroll]

You want to do grid search. The best practice is to try each of those parameters over a range and try all combinations. If you find that your results are improving at the one or another extreme rather in the middle of the range, you'll want to pick a new set of parameters overlapping your margins.

On Fri, Feb 9, 2024 at 7:49 AM Omar Samir <omare...@gmail.com> wrote:

Now I am trying to fine tune Tesseract Arabic model in the testdata_best with more than 20K images in the dataset. I want to know the best values for parameters like MAX_ITERATIONS, EPOCHS, and LEARNING_RATE so how can I find these values I also look inside research papers but I find nothing about the best values for training a model.

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/25365e5b-f591-48d7-8819-124e87be8c0an%40googlegroups.com.

[Omar Samir]

Thank you for that. Another question, is there not any source that suggests parameters for various uses like for training/fine-tuning a model on a specific language or explaining the best values for each use

[Omar Samir]

How can I use grid search to train Tesseract model?