Footnotes extraction for IEEE journals

[q0oz]

Requested feature

...
Hello,

It looks like Docling doesn't recognize footnotes properly for some IEEE manuscripts. For example,
https://arxiv.org/pdf/2503.08661
https://arxiv.org/pdf/2503.08027
https://arxiv.org/pdf/2503.08609

Currently, Docling recognizes those as the main text usually. Please see attached.

Such manuscript templates are quite popular within the IEEE organization and contain important meta-information about the authors.
Would it be possible to retrain/fine-tune the model so that Docling would recognize those as footnotes? Thank you!

Alternatives

I tried to use a simple rule-based approach to check the font size, but unfortunately docling doesn't extract the font size. And it probably wouldn't be that reliable anyway.

[PeterStaar-IBM]

@q0oz Yes, we plan to retrain the layout segmentation. Thank you for providing this sample, it was not on our radar and we definitely need to fix this!

[q0oz]

Thank you!

Just a quick note that you could find more samples, like the ones provided, here: https://arxiv.org/archive/eess in any category.

I also saw a lot of footnotes like that in other IEEE manuscripts which are not publicly available.