InstructDoc: A Dataset for Zero-Shot Generalization of Visual Document Understanding with Instructions

TL;DR

This paper introduces InstructDoc, a large-scale dataset with diverse instructions for visual document understanding, and proposes InstructDr, a model that generalizes well across various VDU tasks and domains using instructions.

Contribution

It provides the first comprehensive VDU dataset with instructions and develops a new instruction-based model that enhances zero-shot generalization capabilities.

Findings

InstructDr outperforms existing multimodal LLMs and ChatGPT on VDU tasks.

The dataset covers 12 tasks and diverse document formats.

The model effectively adapts to new datasets and domains using instructions.

Abstract

We study the problem of completing various visual document understanding (VDU) tasks, e.g., question answering and information extraction, on real-world documents through human-written instructions. To this end, we propose InstructDoc, the first large-scale collection of 30 publicly available VDU datasets, each with diverse instructions in a unified format, which covers a wide range of 12 tasks and includes open document types/formats. Furthermore, to enhance the generalization performance on VDU tasks, we design a new instruction-based document reading and understanding model, InstructDr, that connects document images, image encoders, and large language models (LLMs) through a trainable bridging module. Experiments demonstrate that InstructDr can effectively adapt to new VDU datasets, tasks, and domains via given instructions and outperforms existing multimodal LLMs and ChatGPT without specific training.