On Web-based Visual Corpus Construction for Visual Document Understanding

TL;DR

This paper introduces Webvicob, a web-based tool for creating large-scale, multilingual visual corpora from Wikipedia HTML data, enhancing visual document understanding models especially for resource-scarce languages.

Contribution

Webvicob is a novel dataset generator that constructs extensive visual corpora from raw web data, improving VDU model training and performance on downstream tasks.

Findings

Generated datasets improve VDU model accuracy

Over 13% improvement on DocVQA with Webvicob data

Public implementation available for community use

Abstract

In recent years, research on visual document understanding (VDU) has grown significantly, with a particular emphasis on the development of self-supervised learning methods. However, one of the significant challenges faced in this field is the limited availability of publicly accessible visual corpora or extensive collections of images with detailed text annotations, particularly for non-Latin or resource-scarce languages. To address this challenge, we propose Web-based Visual Corpus Builder (Webvicob), a dataset generator engine capable of constructing large-scale, multilingual visual corpora from raw Wikipedia HTML dumps. Our experiments demonstrate that the data generated by Webvicob can be used to train robust VDU models that perform well on various downstream tasks, such as DocVQA and post-OCR parsing. Furthermore, when using a dataset of 1 million images generated by Webvicob, we observed an improvement of over 13% on the DocVQA Task 3 compared to a dataset of 11 million images from the IIT-CDIP. The implementation of our engine is publicly available on https://github.com/clovaai/webvicob