Getting pre-merged predictions

[hutchhicken]

Hi all. Love the project!

In my use case, I'm trying to get coordinates for all 'Pictures' in a given PDF. I have found that when merge_colliding_predictions is run from get_pdf_segments_for_page in src/vgt/get_most_probable_pdf_segments.py, many images are merge with 'Text' or 'Caption' elements. In my case, I'd prefer to have all the detected elements present, then sort them out / detect collisions on my own.

My quick and dirty hack is simply to comment out line 96 of the file above. Ideally, we'd define this by using a parameter handled by app.py. Is this something other people care about (i.e., is worth my submitting a PR), or is this just me?

[gabriel-piles]

Thank you for your interest in this project.

It's possible that more people are interested in this feature, so it's worth giving it a try. We have other priorities right now, but maybe we can do it in the future. However, if you are interested, you can create a PR with the development.

Our thinking is to avoid extra parameters and add this information to a new property in the response object (SegmentBox). This property could be called sub_elements_positions or something similar.

Have a good day.