Lumos : Empowering Multimodal Llms With Scene Text Recognition

Shenoy Ashish, Lu Yichao, Jayakumar Srihari, Chatterjee Debojeet, Moslehpour Mohsen, Chuang Pierce, Harpale Abhay, Bhardwaj Vikas, Xu Di, Zhao Shicong, Zhao Longfang, Ramchandani Ankit, Dong Xin Luna, Kumar Anuj. Arxiv 2024

[Paper]    
Efficiency And Optimization Model Architecture Multimodal Models TACL

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.

Similar Work