DDI-100
收藏资源简介:
DDI-100是由莫斯科物理技术学院创建的一个大型合成数据集,基于7000个真实文档页面生成了超过100000张经过增强的图像。数据集包含文本和印章的掩码、文本和字符的边界框及相关标注。该数据集通过多种文本检测和光学字符识别模型验证,显示出高质量的性能。DDI-100适用于文本检测、光学字符识别和印章检测等文档图像分析领域,旨在解决现有数据集规模小、难以比较模型性能的问题。
DDI-100 is a large-scale synthetic dataset developed by the Moscow Institute of Physics and Technology. It generates over 100,000 augmented images based on 7,000 real document pages. This dataset includes masks for text and seals, bounding boxes for text and characters, along with relevant annotations. It has been validated by multiple text detection and optical character recognition (OCR) models, demonstrating high-quality performance. DDI-100 is applicable to document image analysis tasks such as text detection, OCR and seal detection, and aims to address the issues of small-scale existing datasets and the difficulty in comparing model performance.




