doc-analysis / xfund Goto Github PK

Excuse me, 1 What tools are used to label your xfund dataset. 2. What tools are used to mark the data set of funsd

In your article you describe relation extraction task. In this task you require as an output set of tuples of three in form of
(head_entity, tail_entity, relation_label). However, in the dataset you publish one may find only relations in form defined by documents/*/document/*/linking, which allows to create tuples of two (head_entity, tail_entity). Information about relation label is missing. Is it present somewhere else in your dataset, or it was by some mistake omitted when the dataset was published?

hi, I wanted to know that, why zh and ja datasets are split by character? not word by word?
when building a dataset, sentences can be split by words, not characters?
thank you.

Dear all, I am trying to train layoutLM Relation extraction model with FUNSD, as the RE model was pre-trained on XFUN, so I need to convert FUNSD to XFUN format, thus I would like to ask if anyone has any idea on how to convert the FUNSD to XFUN? Very much appreciated your help!

Ps: I have tried to build the conversion script, but am a bit confused with the derivation of word level bboxes, in another word, how can we derive the word level bboxes based on FUNSD? (because the texts in FUNSD is entity level instead of word level)

Hi, may I ask if the annotation rule is the same as FUNSD? There are a number of strange examples.

Is there any way to validate on english dataset?

Hi

I am new to the LayoutXLM model. I have taken the dataset from https://github.com/doc-analysis/XFUND/releases/tag/v1.0
these files de.train.json,de.train.zip,de.val.json,de,val.zip and uploaded to the https://github.com/jyotiyadav94/Relational-Extraction/releases/tag/xfund
whenever I try to access using the below script where the only change is URL I am getting JSONDecode error every time.
I am bit confused why i am always getting this although the dataset is the same.

Hi,

I found that ZH's stats are different from the paper ones, but exactly the same as FUNSD's, is there a mistake?

[Screenshot from this repo]

[Screenshot from LayoutXLM paper]

[Screenshot from FUNSD paper]