Inconsistent Table data extraction (same page, different documents)

[jeromemassot]

Hi all,

I am experiencing a strange issue here: the Deep Seach toolkit is not able to format consistently the content extracted from the table. In fact, the table['cells']['data'] contains extracted features but the table['data'] is empty. I can understand this behavior when the table layout is complex but for easy layout, there is no explanation.

The stranger is the fact that this issue is not reproducible. When I tried to debug the issue with a pdf file containing the single page where the table is located, this time the table['data'] field is populated with the content. So it looks like the table layout understanding is also linked to the length of the pdf file, more precisely the page numbers.

Could it be because of a memory overflow somewhere? Or because there is some memory limit reached?
Several tables located after the faulty table in the original document are extracted correctly...

I attach here the both pdf files and the JSON files received from the toolkit.

Access to files here.

In the long JSON file, the issue is located at line 12317, which corresponds to line 375 in the short JSON file.

Thanks in advance for investigating this issue which is causing major damage in my application.

Best regards

Jerome

[staar]

@jeromemassot Let us have a quick look! We have updated our table models (see here).

Thank you for bringing this up!