# General OCR Theory: Towards OCR-2.0 via a Unified End-to-end Model

**Authors:** Haoran Wei, Chenglong Liu, Jinyue Chen, Jia Wang, Lingyu Kong, Yanming Xu, Zheng Ge, Liang Zhao, Jianjian Sun, Yuang Peng, Chunrui Han, Xiangyu Zhang  
**arXiv:** [2409.01704](https://arxiv.org/abs/2409.01704)  
**Date:** 2024-09-03  
**Summary page:** https://www.summarizepaper.com/en/paper/2409.01704v1/

## Key points

- Authors propose the General OCR Theory (GOT) and model for advancing OCR to OCR-2.0
- GOT model designed to handle various types of artificial optical signals, including plain texts, formulas, tables, charts, sheet music, and shapes
- GOT model has 580M parameters, high-compression encoder, long-contexts decoder for superior performance
- Authors gathered diverse data sources using six rendering tools like LATEX and Mathpix-markdown-it for training and testing
- Model's capabilities enhanced through challenging tasks like sheet music recognition and chart analysis using datasets like GrandStaff
- Comprehensive approach to data collection and model development demonstrates effectiveness and versatility of GOT model in handling OCR tasks with superior performance

## Summary

The authors propose the General OCR Theory (GOT) and an associated model to advance the field of Optical Character Recognition (OCR) from traditional systems to OCR-2.0. The GOT model is designed to handle various types of artificial optical signals, or "characters," including plain texts, math/molecular formulas, tables, charts, sheet music, and geometric shapes. With 580M parameters and a high-compression encoder and long-contexts decoder for end-to-end processing, the GOT model offers superior performance compared to traditional OCR systems. To enhance its capabilities, the authors have gathered diverse data sources for training and testing purposes using six rendering tools such as LATEX for tables and Mathpix-markdown-it for math/molecular formulas. They also aim to expand the model's capabilities by collecting challenging tasks like sheet music recognition and chart analysis through datasets like GrandStaff. Through their comprehensive approach to data collection and model development, the authors demonstrate the effectiveness and versatility of their GOT model in handling a wide range of optical character recognition tasks with superior performance compared to traditional OCR systems.