Exploring the Capabilities of Large Multimodal Models on Dense Text
This work addresses the need for better understanding of LMM capabilities in dense text applications, such as document analysis, but it is incremental as it focuses on evaluation and dataset creation rather than novel model development.
The authors tackled the problem of evaluating large multimodal models (LMMs) on dense text tasks by introducing the DT-VQA dataset with 170k question-answer pairs, revealing strengths and weaknesses of models like GPT4V and Gemini, and showing that prompt engineering and fine-tuning lead to significant performance improvements.
While large multi-modal models (LMM) have shown notable progress in multi-modal tasks, their capabilities in tasks involving dense textual content remains to be fully explored. Dense text, which carries important information, is often found in documents, tables, and product descriptions. Understanding dense text enables us to obtain more accurate information, assisting in making better decisions. To further explore the capabilities of LMM in complex text tasks, we propose the DT-VQA dataset, with 170k question-answer pairs. In this paper, we conduct a comprehensive evaluation of GPT4V, Gemini, and various open-source LMMs on our dataset, revealing their strengths and weaknesses. Furthermore, we evaluate the effectiveness of two strategies for LMM: prompt engineering and downstream fine-tuning. We find that even with automatically labeled training datasets, significant improvements in model performance can be achieved. We hope that this research will promote the study of LMM in dense text tasks. Code will be released at https://github.com/Yuliang-Liu/MultimodalOCR.