IDPL-PFOD2

IDPL-PFOD2是由智能数据处理实验室开发的针对印刷体波斯语光学字符识别的大型数据集，包含2,003,541张图像，涵盖多种字体、风格和大小。该数据集是对先前IDPL-PFOD数据集的扩展，显著增加了数据量和多样性。IDPL-PFOD2旨在解决波斯语印刷文本识别的挑战，特别是由于波斯语独特的书写特征和深度学习架构对大量训练样本的需求。数据集通过CRNN和Vision Transformer架构的有效性评估，展示了其在波斯语OCR研究中的应用潜力，为波斯语社区提供更好的可访问性、信息检索和语言处理能力。IDPL-PFOD2 is a large-scale dataset developed by the Intelligent Data Processing Laboratory for printed Persian optical character recognition (OCR). It contains 2,003,541 images covering diverse fonts, styles and sizes. As an extension of the previously released IDPL-PFOD dataset, it has significantly expanded its data volume and diversity. IDPL-PFOD2 aims to address the challenges in Persian printed text recognition, particularly those arising from the unique orthographic features of the Persian language and the large-scale training sample requirements of deep learning architectures. The effectiveness of the dataset has been evaluated using CRNN and Vision Transformer architectures, demonstrating its application potential in Persian OCR research and providing better accessibility, information retrieval and language processing capabilities for the Persian-speaking community.

搜集汇总

背景与挑战

背景概述

IDPL-PFOD2是一个大规模印刷波斯语文本图像数据集，专为波斯语光学字符识别（OCR）研究设计。该数据集包含超过200万张PNG格式图像，每张图像尺寸为300*50像素，代表一行真实波斯语文本，并分为训练集、验证集和测试集提供下载。数据集基于人工生成，旨在支持波斯语OCR技术和自然语言处理应用的发展。