DocVQA
收藏资源简介:
文档视觉问答 (DocVQA) 旨在激发文档分析和识别研究中的 “目的驱动” 观点,其中文档内容被提取并用于响应由此信息的人类消费者定义的高级任务。为此,我们组织了一系列挑战并发布数据集,以使机器能够 “理解” 文档图像,从而回答有关它们的问题。 数据集中有50 k个问题和12k个图像。图像是从UCSF行业文档库中收集的。问题和答案是手动注释的。
Document Visual Question Answering (DocVQA) aims to inspire a "purpose-driven" perspective in document analysis and recognition research, where document content is extracted and utilized to address high-level tasks defined by human consumers of such information. To this end, we have organized a series of challenges and released the dataset to enable machines to "understand" document images and thus answer questions about them. The dataset contains 50,000 questions and 12,000 images. The images were collected from the UCSF Industry Document Library. The questions and answers were manually annotated.




