We introduce Lumos, the first end-to-end multimodal question-answering system with text understanding capabilities. At the core of Lumos is a Scene Text Recognition (STR) component that extracts text from first person point-of-view images, the output of which is used to augment input to a Multimodal Large Language Model (MM-LLM). While building Lumos, we encountered numerous challenges related to STR quality, overall latency, and model inference. In this paper, we delve into those challenges, and discuss the system architecture, design choices, and modeling techniques employed to overcome these obstacles. We also provide a comprehensive evaluation for each component, showcasing high quality and efficiency.
@article{arxiv.2402.08017,
title = {Lumos : Empowering Multimodal LLMs with Scene Text Recognition},
author = {Ashish Shenoy and Yichao Lu and Srihari Jayakumar and Debojeet Chatterjee and Mohsen Moslehpour and Pierce Chuang and Abhay Harpale and Vikas Bhardwaj and Di Xu and Shicong Zhao and Longfang Zhao and Ankit Ramchandani and Xin Luna Dong and Anuj Kumar},
journal= {arXiv preprint arXiv:2402.08017},
year = {2024}
}