JSON syntax for elements located in several bounding boxes

[bdoubrov]

Currently location of any content node is defined as a single bounding box on a page (bounding box and page number properties). This is not good enough if an element is spread across several pages or even several columns on the same page. This is a typical use case for paragraphs, list and list items, table rows and tables.

[MaximPlusov]

The current version:

    "page number" : 1,
    "bounding box" : [ 215.57, 710.987, 480.851, 727.009 ]

The possible option without backward compatibility:

"location": [{
    "page number" : 1,
    "bounding box" : [ 215.57, 710.987, 480.851, 727.009 ]
},
{
    "page number" : 2,
    "bounding box" : [ 106.827, 461.693, 505.173, 496.698 ]
}
]

[bdoubrov]

This is not the most common case. I would suggest to try preserving backward compatibility. Or at least accept both versions of defining the location.

[denisbialy]

One element that stays within one page should have one bounding box. I'm not sure that I see the use case of having one element with multiple bounding boxes on the same page. If it crosses some columns, then it will either be split by them into multiple elements, or those columns are joined, and there should be one bounding box.

For maximum backward compatibility we can utilize "id" field that every table has. Make ids unique, add new field "next_table" or something like that. This will allow us to merge multiple tables into one multi-page table at later steps of processing and is 100% optional and backward compatible. Probably the easiest option for us as well.