VR_ClassifiedDataset

VR_ClassifiedDataset是由大阪大学创建的一个视觉关系真实性数据集，包含31,769条数据，旨在区分图像中对象间的真实与虚假关系。该数据集基于Flickr30k实体图像标题数据集，通过依赖解析和众包标注方法构建，涵盖了所有在标题中出现的对象间关系。数据集不仅支持图像理解研究，还对基于视觉的自然语言处理任务有重要贡献，如视觉基础的释义识别和名词短语的依赖解析。

VR_ClassifiedDataset is a visual relationship authenticity dataset developed by Osaka University, which contains 31,769 data instances. It is designed to discriminate between true and false relational connections between objects in images. Built upon the Flickr30k Entities image caption dataset via dependency parsing and crowdsourced annotation methods, this dataset covers all inter-object relationships present in the image captions. Furthermore, it not only supports image understanding research, but also makes significant contributions to vision-based natural language processing tasks, such as vision-grounded paraphrase recognition and dependency parsing of noun phrases.