PDF annotation: tables, figures, formulas should be excluded

[lfoppiano]

I've noticed that tables and figures, with the current amount of training data, are incorrectly annotated. Document hal-00643787.

See example of a figure:

here an example on a table:

here a formula:

[kermitt2]

yes I was a bit in a hurry so I didn't go fine-grained for processing the PDF via GROBID. Right now the quantity parser is applied on the title, abstract, the whole full text and the annexes (the rest is ignored).

So to be done : in the body part, we should exclude formula, all the ref markers (ref of biblio [Foppiano and al. 2017], figure, formula, etc.) present in the body and figures/tables for the moment.

[lfoppiano]

OK, so I excluded <formula>, <table>, <equation> and <figure>

[lfoppiano]

here how it looks a figure:

[kermitt2]

Nice!
Maybe we could also let the model annotate the caption and figure/table titles?

Then it requires the figure and table model to be applied to the LayoutTokens labelled with TaggingLabels.FIGURE and TaggingLabels.TABLE

[lfoppiano]

Now only captions are processed:

[lfoppiano]

I think this is solved. Sometimes tables are annotated anyway, but this is more due to the table model.