cmarkea

/

detr-layout-detection

like 25

Follow

Credit Mutuel Arkea 64

Hi,

I tried the model, and I take it a good job. But there is still remaining one problem left, that is the list-items and footnotes are really confused.

I tried to review the dataset, and found that footnotes are less than 1% percent of all the annotations which is not very much. Besides, list items and footnotes have similar characteristics, but their key difference is their placement on the page. So, would it be possible to include an element's positional data as another input dimension for the model?

Are there any existing efforts to implement this adjustment?