I found some bug with the reading order for the layout plugin in the attached pdf on page 2(also in the online demo).
Hope this will be fixed in the future, keep on the great work!
9.pdf (3,3 MB)
@dogo Welcome!
Confirming your observation:
The line starting with “of novel content (Pretorius,…” was incorrectly classified as a separate “text” layout box. It should have been a part of the text paragraph above.
Because the Layout plugin is AI-based, its decisions are results of plausibility calculations - and thus not necessarily logical in each single case.
Here, this has confused the reading order detector.
We are currently training a new generation of Layout detector models. Let’s evaluate the expected improvements before we consider other solutions.