The FUNSD_polygon dataset has the following properties:
{
"form":[
{
"text":"12 /10 /98",
"linking":[
[
2,
27
]
],
"label":"answer",
"words":[
{
"text":"12",
"polygon":[
184,
406,
198,
406,
198,
420,
184,
420
]
},
{
"text":"/10",
"polygon":[
198,
405,
216,
405,
216,
423,
198,
423
]
},
{
"text":"/98",
"polygon":[
215,
406,
233,
406,
233,
423,
215,
423
]
}
],
"id":27,
"polygon":[
184,
405,
233,
405,
233,
423,
184,
423
]
}
]
}
Where each property has the following meaning:
form
: The form that contains the text.text
: The text of a given phrase.linking
: The linking of an answer to a question. EG "DATE: 12/10/98"label
: The label of the text. EG "answer"words
: The words that compose the text.id
: The id of the text.polygon
: The polygon that contains the text in the format[xleft, ytop, xright, ytop, xright, ybottom, xleft, ybottom]
words.text
: The text of a given word.words.polygon
: The polygon that contains the word in the format[xleft, ytop, xright, ytop, xright, ybottom, xleft, ybottom]
For more information on the dataset, please refer to the original paper.