General

[ghost]

Discussing general topic.

[lquirosd]

Topic: suggestion

@lquirosd You need to merge the pre_trained/README.md with the master/README.md so it's easier to find for new members.

there is a link on master/README.md to pre_trained/README.md, there is a lot of work to be done related to pre-trained models, both README files should be updated soon.

[lquirosd]

Topic: question

• Was ALAR_min_model_17_12_18.pth trained to detect base-lines or text-lines?

That model is trained to extract baselines only, but text-lines can be obtained easily from there

• How to choose the right configurations to train to detect Text-Lines?
  I have seen the config help, but what do you recommend?

This software is not intended to extract text-lines directly, instead we rely on baselines (so users can easily fix any error) and then another software can be used to extract the text-lines from the baselines.

I recommend this software for that task pageContourGenerator to generate contours around the baselines (bases on the inter-line distance) and pageLineExtractor to extract those contours from the original image (i.e. generate a set of sub-images from the original one).