ICDAR 2023 Competition on Document UnderstanDing of Everything (DUDE).
ICDAR (2)(2023)
Abstract
This paper presents the results of the ICDAR 2023 competition on Document UnderstanDing of Everything. DUDE introduces a new dataset comprising 5 K visually-rich documents (VRDs) with 40 K questions with novelties related to types of questions, answers, and document layouts based on multi-industry, multi-domain, and multi-page VRDs of various origins and dates. The competition was structured as a single task with a multi-phased evaluation protocol that assesses the few-shot capabilities of models by testing generalization to previously unseen questions and domains, a condition essential to business use cases prevailing in the field. A new and independent diagnostic test set is additionally constructed for fine-grained performance analysis. A thorough analysis of results from different participant methods is presented. Under the newly studied settings, current state-of-the-art models show a significant performance gap, even when improving visual evidence and handling multi-page documents. We conclude that the DUDE dataset proposed in this competition will be an essential, long-standing benchmark to further explore for achieving improved generalization and adaptation under low-resource fine-tuning, as desired in the real world.
MoreTranslated text
Key words
document understanding
AI Read Science
Must-Reading Tree
Example
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined