Saved in:
Bibliographic Details
Main Authors: Uthayasooriyar, Benno, Ly, Antoine, Vermet, Franck, Corro, Caio
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.08606
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918299186495488
author Uthayasooriyar, Benno
Ly, Antoine
Vermet, Franck
Corro, Caio
author_facet Uthayasooriyar, Benno
Ly, Antoine
Vermet, Franck
Corro, Caio
contents We introduce DocPolarBERT, a layout-aware BERT model for document understanding that eliminates the need for absolute 2D positional embeddings. We extend self-attention to take into account text block positions in relative polar coordinate system rather than the Cartesian one. Despite being pre-trained on a dataset more than six times smaller than the widely used IIT-CDIP corpus, DocPolarBERT achieves state-of-the-art results. These results demonstrate that a carefully designed attention mechanism can compensate for reduced pre-training data, offering an efficient and effective alternative for document understanding.
format Preprint
id arxiv_https___arxiv_org_abs_2507_08606
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DocPolarBERT: A Pre-trained Model for Document Understanding with Relative Polar Coordinate Encoding of Layout Structures
Uthayasooriyar, Benno
Ly, Antoine
Vermet, Franck
Corro, Caio
Computation and Language
We introduce DocPolarBERT, a layout-aware BERT model for document understanding that eliminates the need for absolute 2D positional embeddings. We extend self-attention to take into account text block positions in relative polar coordinate system rather than the Cartesian one. Despite being pre-trained on a dataset more than six times smaller than the widely used IIT-CDIP corpus, DocPolarBERT achieves state-of-the-art results. These results demonstrate that a carefully designed attention mechanism can compensate for reduced pre-training data, offering an efficient and effective alternative for document understanding.
title DocPolarBERT: A Pre-trained Model for Document Understanding with Relative Polar Coordinate Encoding of Layout Structures
topic Computation and Language
url https://arxiv.org/abs/2507.08606