数据堂—222,289张中文自然场景OCR数据

222,289张中文自然场景OCR数据采集场景涵盖室内和室外场景。数据多样性包括多种场景和多种拍摄角度。在标注方面，对图像中文字进行行级标注、单词级标注、字符级标注和内容转写。本套中文自然场景OCR数据可用于自然场景OCR任务。

This dataset includes 222,289 Chinese natural scene OCR samples. Its data collection scenarios cover both indoor and outdoor environments. The data features diversity across multiple scenarios and diverse shooting angles. For annotation, text within the images is annotated at line-level, word-level and character-level, with complete content transcription performed. This Chinese natural scene OCR dataset can be applied to natural scene OCR tasks.

搜集汇总

背景与挑战

背景概述

该数据集包含222,289张中文自然场景图像，覆盖室内外多种场景和拍摄角度，用于自然场景OCR测试任务。图像标注包括行、词、字符级别的矩形框和内容转录，准确率不低于95%，数据格式为图像文件（如jpg、png）和json标注文件。