BoundingDocs
收藏资源简介:
BoundingDocs是一个由佛罗伦萨大学和LETXBE联合创建的文档问答数据集,旨在解决文档AI领域中的信息提取和视觉问答任务。该数据集整合了多个公开数据集,涵盖了丰富的文档类型和语言特征,提供了OCR文本和答案在文档图像中的精确位置信息。数据集通过统一现有数据集并增强布局注释,生成了适用于训练和评估大型语言模型的问答格式。数据集的应用领域包括文档理解、信息提取和视觉问答,旨在通过提供精确的空间坐标信息,减少模型幻觉并提升文档布局理解的准确性。
BoundingDocs is a document question answering dataset jointly created by the University of Florence and LETXBE, aimed at addressing information extraction and visual question answering tasks in the field of document AI. This dataset integrates multiple public datasets, covers diverse document types and rich linguistic characteristics, and provides precise spatial position information of OCR texts and answers within document images. By unifying existing datasets and augmenting layout annotations, the dataset constructs question-answering formats suitable for training and evaluating large language models. The application scenarios of this dataset cover document understanding, information extraction and visual question answering, and it aims to reduce model hallucinations and improve the accuracy of document layout understanding by providing precise spatial coordinate information.




