Benchmarking Chinese Text Recognition
收藏资源简介:
本研究构建了一个名为‘Benchmarking Chinese Text Recognition’的综合数据集,旨在推动中文文本识别技术的发展。数据集由复旦大学智能信息处理上海市重点实验室团队收集,涵盖了场景、网页、文档和手写四大类别,总计636,455条数据。数据集通过公开竞赛、项目和论文中的数据手动收集,并根据应用场景进行了分类。此外,研究团队还标准化了中文文本识别的评估协议,并在统一的评估标准下,对一系列代表性文本识别方法进行了评估,以提供基准性能。数据集的应用领域广泛,包括自动驾驶、文档检索和签名识别等,旨在解决中文文本识别领域的关键问题,如缺乏合理的标准和统一的评估协议。
This study constructs a comprehensive dataset named 'Benchmarking Chinese Text Recognition' to advance the development of Chinese text recognition technologies. This dataset was collected by the team from the Shanghai Key Laboratory of Intelligent Information Processing, Fudan University, covering four major categories: scene, web, document, and handwritten, with a total of 636,455 data entries. The dataset was manually collected from publicly available data in competitions, projects, and academic papers, and classified according to application scenarios. Additionally, the research team standardized the evaluation protocols for Chinese text recognition, and evaluated a series of representative text recognition methods under the unified evaluation criteria to provide benchmark performance. The dataset has wide application scenarios, including autonomous driving, document retrieval, signature recognition and other fields, aiming to address the key challenges in the field of Chinese text recognition, such as the lack of reasonable standards and unified evaluation protocols.

- 1Benchmarking Chinese Text Recognition: Datasets, Baselines, and an Empirical Study复旦大学智能信息处理上海市重点实验室 · 2022年



