PDF to Markdown conversion of "Attention is all you need" paper's pdf from arxiv is having a bug

[amal5haji]

Bug

Converting "Attention is all you need" paper to docling is giving corrupted response for the images at the end of the PDF, at the end of this paper there are few images on a section called "Attention Visualizations" all this images in this section is not being converted by docling for some reaason. please look into it

Steps to reproduce

Download and just convert this pdf to see the issue

Docling version

latest

[dosubot]

Hi @amal5haji! I'm Dosu and I’m helping the docling team.

This issue happens because Docling's layout postprocessing filters out images (PICTURE clusters) that cover more than 90% of the page area, which is common for the large visualizations at the end of the "Attention is all you need" paper. As a result, these images are omitted from the conversion output—even though they're detected—due to this hardcoded area threshold in the postprocessor logic. There isn't any special handling for images at the end of documents; the omission is purely due to this area-based rule. To fix this, you would need to adjust or parameterize the area filter for image clusters in the layout postprocessor code. For reference, see the relevant logic in layout_postprocessor.py.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}