Hi, I found a small regression with the latest version of grobid, for what concern the results of the segmentation model with the new pdfalto version (0.5).

First of all, I just realised that the pdf alto version for mac was not updated (still at version 0.3) on master. Should I update to the latest one?

The segmentation model seems to behave differently including a large part of the body in the header. In particular, the publication date is not correctly extracted.

Here an example paper: https://aip.scitation.org/doi/pdf/10.1063/1.2925452

Here the output from grobid:

• Fulltext extraction (pdfalto 0.5) 1.2925452_processPDF_pdfalto0.5.pdf.tei.xml.zip
• Header extraction (pdflato 0.5) 1.2925452_processHeader_pdfalto0.5.pdf.tei.xml.zip

Here the result from the segmentation model:

• Result of the segmentation model, ran on a docker image, pdfalto 0.5
• Result of the segmentation model, run natively on my mbp, pdflato 0.3