数据堂—20,011张自然场景OCR描述数据

20,000张自然场景OCR描述数据，包含亚洲语系与欧洲语系，总计14种语言，采集环境包括商店牌匾、站牌、海报、路标等多种场景，含有多种拍摄角度。描述语言为英语，主要描述文字排列方式、文字内容，颜色等信息

This dataset comprises 20,000 natural scene OCR description samples, covering 14 languages across both Asian and European language families. The data was collected from diverse scenarios including storefront signs, bus stop signs, posters, road signs and other similar public display settings, with multiple shooting angles involved. All descriptive annotations are in English, focusing on details such as text layout, text content and color information of the captured scenes.

搜集汇总

背景与挑战

背景概述

该数据集包含20,000张自然场景图像及对应英文描述，用于OCR描述测试任务，覆盖14种亚洲和欧洲语言，采集自店铺招牌、车站标志等多种环境。描述内容主要涉及文本排列、内容、颜色和场景信息，标注准确率不低于97%。