500,000张21国自然场景&文档拍照场景&电子场景OCR标注数据

500,000张21国自然场景&文档拍照场景&电子场景OCR标注数据。数据包含21个语种，每个语种数量分布为20,000-25,000张。数据类型包括自然场景、文档拍照场景和电子场景。数据多样性包括多种数据类型、多种拍摄角度、多语种。在标注方面，采用行（列）级四边形或多边形标注、行（列）级内容转写。数据可用于多语种OCR识别任务。

This dataset comprises 500,000 OCR-annotated images covering 21 languages, including three scenario types: natural scenes, document photography scenarios, and electronic scenarios. It distributes 20,000 to 25,000 images per language. The data features rich diversity across multiple data types, shooting angles, and language varieties. For annotation, row-level or column-level quadrilateral or polygonal bounding boxes are adopted, alongside transcription of the content corresponding to each row or column. This dataset can be applied to multilingual OCR recognition tasks.

搜集汇总

背景与挑战

背景概述

该数据集包含50万张图像，覆盖21种语言，涵盖自然场景、文档拍摄和电子场景，并标注了行级或列级的文本边界框及转录内容。它适用于多语言OCR识别任务，数据来源多样且标注准确度高。