OCR or Not? Rethinking Document Information Extraction in the MLLMs Era with Real-World Large-Scale Datasets

TL;DR

This study benchmarks multimodal large language models on document extraction tasks, revealing that image-only inputs can match OCR+MLLM performance, and provides insights into optimizing MLLMs for real-world applications.

Contribution

It introduces a large-scale benchmarking framework and an automated error analysis method, challenging the necessity of OCR in MLLMs for document extraction.

Findings

Image-only input achieves comparable performance to OCR-enhanced methods.

Careful schema, exemplars, and instructions improve MLLMs performance.

Automated error analysis reveals key failure modes.

Abstract

Multimodal Large Language Models (MLLMs) enhance the potential of natural language processing. However, their actual impact on document information extraction remains unclear. In particular, it is unclear whether an MLLM-only pipeline--while simpler--can truly match the performance of traditional OCR+MLLM setups. In this paper, we conduct a large-scale benchmarking study that evaluates various out-of-the-box MLLMs on business-document information extraction. To examine and explore failure modes, we propose an automated hierarchical error analysis framework that leverages large language models (LLMs) to diagnose error patterns systematically. Our findings suggest that OCR may not be necessary for powerful MLLMs, as image-only input can achieve comparable performance to OCR-enhanced approaches. Moreover, we demonstrate that carefully designed schema, exemplars, and instructions can further enhance MLLMs performance. We hope this work can offer practical guidance and valuable insight for advancing document information extraction.