Vsualized region of figure or table by pdfbox

[Sunnycheey]

Is there some demo code for marking the content out by the coordinate (i.e, p, x, y, w, h) with pdfbox, since I really don't want to go deep with pdf....

Any help will be appreciated.

[kermitt2]

Hello @Sunnycheey

Yes you can have a look at the class FigureTableVisualizer.java under grobid/grobid-core/src/main/java/org/grobid/core/visualization/. The other classes in the package are good examples too of usage of pdfbox 2.* for extra-annotations with the Grobid results.

[ehapmgs]

Any luck in doing so?

I believe it should look something like this based on the code sample

// grobid coords
const [page, x, y, w, h] = [6, 177.28, 358.04, 305.92, 8.42]
// page media box 
const mediaBox = { lowerX: 0, lowerY: 0, width: 667.28, height: 913.89 }
const annX = x + mediaBox.lowerX
const annY = mediaBox.height - (y + h) + mediaBox.lowerY
const annRightX = x + w + mediaBox.lowerY
const annTopY = mediaBox.height - y + mediaBox.lowerY
console.log(annX, annY, annRightX , annTopY)

but it totally points to something else in my case

[kermitt2]

@ehapmgs hello!

Do you want to annotate a PDF with pdfbox (so producing a new PDF with annotations) or annotate a PDF with PDF.js in a browser?

If first, you have examples under grobid/grobid-core/src/main/java/org/grobid/core/visualization/ as indicated above.

If second, you are in the wrong issue, but look at the web demo:
https://github.com/kermitt2/grobid/blob/master/grobid-service/src/main/resources/web/grobid/grobid.js#L332 (PDF display)
https://github.com/kermitt2/grobid/blob/master/grobid-service/src/main/resources/web/grobid/grobid.js#L646 (annotation layer on top of the PDF)

Basically you need to scale the annotations (which have no unit) according to the canvas where the PDF is displayed.

                var page = thePos.p;
		var pageDiv = $('#page-'+page);
		var canvas = pageDiv.children('canvas').eq(0);;

		var canvasHeight = canvas.height();
		var canvasWidth = canvas.width();
		var scale_x = canvasHeight / page_height;
		var scale_y = canvasWidth / page_width;

		var x = thePos.x * scale_x;
		var y = thePos.y * scale_y;
		var width = thePos.w * scale_x;
		var height = thePos.h * scale_y;

[ehapmgs]

Hey @kermitt2

Thanks for the help yes it is the first case but it seems there is something wrong with the coords probably a bug that I need to open a new issue for.

Using the example in grobid/grobid-core/src/main/java/org/grobid/core/visualization/ you can see the annotations are not located correctly

I believe it used to work in version 0.5.4 I can see that the coords for the same figure is different between the master branch and 0.5.4

[kermitt2]

Indeed thanks ! I didn't look at this part since very long, I am re-discovering it :)

There are actually web services to get the annotated PDF:

curl -v --form input=@./s41523-020-00198-1.pdf --form type=2 localhost:8070/api/annotatePDF > s41523-020-00198-1-annot.pdf

s41523-020-00198-1-annot.pdf gives the PDF with annotations, the attribute type indicates which annotations to put (0: citations, 1: blocks, 2: figure).

This is working on some PDF:

But for the one you tested, there is a scaling problem. I think Achraf fixed this problem in pdfalto with kermitt2/pdfalto#43 - see #330

So just need to do the same with pdfbox here... mmm

[kermitt2]

For reference, failing PDF https://onlinelibrary.wiley.com/doi/pdf/10.1087/20100308

[kermitt2]

Normally fixed with 33b50b7

curl -v --form input=@./20100308.pdf --form type=2 localhost:8070/api/annotatePDF > 20100308-annot.pdf

[ehapmgs]

@kermitt2 nice! it is working as expected now thanks