IndicSTR12

IndicSTR12是一个专为印度场景文本识别设计的大型真实数据集，由国际信息技术研究所视觉信息中心创建。该数据集包含超过27000个从各种自然场景中收集的单词图像，每种语言至少有1000个单词图像。数据集的创建过程涉及从Google Images爬取图像，并进行了详细的标注，包括四角点标注和分类。IndicSTR12旨在解决印度语言在场景文本识别领域的数据稀缺问题，支持多种印度语言，包括Assamese, Bengali, Odia, Marathi, Hindi, Kannada, Urdu, Telugu, Malayalam, Tamil, Gujarati, Punjabi等。数据集的应用领域广泛，包括图像搜索、翻译、辅助技术等，特别适用于解决多语言环境下的文本识别问题。

IndicSTR12 is a large-scale real-world dataset tailored for Indian scene text recognition, developed by the Visual Information Center at the International Institute of Information Technology. It comprises over 27,000 word images harvested from diverse natural scenes, with no fewer than 1,000 word images per language. The dataset was constructed via web crawling of images from Google Images, followed by rigorous annotations encompassing quadrilateral corner point labeling and language classification. Designed to mitigate the data scarcity challenge in Indian language scene text recognition, IndicSTR12 supports a wide range of Indian languages, namely Assamese, Bengali, Odia, Marathi, Hindi, Kannada, Urdu, Telugu, Malayalam, Tamil, Gujarati, and Punjabi. With broad application scenarios including image search, machine translation, and assistive technology, this dataset is particularly well-suited for addressing text recognition tasks in multilingual settings.