caoxuhao/SARD

SARD（合成阿拉伯语识别数据集）是一个大规模、合成生成的数据集，专为阿拉伯文本的光学字符识别（OCR）模型训练和评估而设计。该数据集通过提供模拟真实世界书籍布局的受控、多样化和可扩展的训练数据，解决了对全面阿拉伯语文本识别资源的关键需求。数据集具有大规模（2,515,082张文档图像，包含7.946亿单词）、字体多样性（六种不同的阿拉伯字体：Amiri、Sakkal Majalla、Arial、Calibri、Scheherazade New和Traditional Arabic）、结构化格式（模拟书籍布局）、干净数据（无扫描伪影、模糊或失真）和内容多样性（涵盖文化、文学、伊斯兰教法、社会等多个领域）等关键特征。数据集按字体名称分为六个部分，每个部分包含图像名称、文本块、字体名称、Base64编码的图像表示、样本ID和源文章链接等属性。

SARD (Synthetic Arabic Recognition Dataset) is a large-scale, synthetically generated dataset designed for training and evaluating Optical Character Recognition (OCR) models for Arabic text. This dataset addresses the critical need for comprehensive Arabic text recognition resources by providing controlled, diverse, and scalable training data that simulates real-world book layouts. Key features include massive scale (2,515,082 document images containing 794.6 million words), typographic diversity (six distinct Arabic fonts: Amiri, Sakkal Majalla, Arial, Calibri, Scheherazade New, and Traditional Arabic), structured formatting (mimicking book layouts), clean data (no scanning artifacts, blur, or distortions), and content diversity (spanning multiple domains such as culture, literature, Shariah, social topics, and more). The dataset is divided into six splits based on font name, each with attributes like image name, text chunk, font name, Base64-encoded image representation, sample ID, and article link.