OpenEQA
收藏资源简介:
OpenEQA是由乔治亚理工学院创建的体感式问答数据集,旨在推动基于基础模型的具身智能研究。该数据集包含从ScanNet和HM3D环境中采集的具身交互式问答对,涵盖对象识别、属性识别、空间理解等多个类别。数据集共包含63个HM3D验证环境的具身历史记录,以及由8位AI研究人员标注的问答对。其创建过程通过半自动化方式生成具身历史,并由人工验证问答对的准确性。OpenEQA主要用于评估具身智能体在复杂环境中的问答能力,助力研究者探索视觉感知与语言理解的结合。
OpenEQA is an embodied question answering dataset created by the Georgia Institute of Technology, aiming to advance embodied intelligence research based on foundation models. This dataset contains embodied interactive question-answering pairs collected from ScanNet and HM3D environments, covering multiple categories such as object recognition, attribute recognition and spatial understanding. It includes embodied historical records from 63 HM3D validation environments, as well as question-answering pairs annotated by 8 AI researchers. Its creation process adopts a semi-automated approach to generate embodied histories, with manual verification conducted for the accuracy of the question-answering pairs. OpenEQA is primarily used to evaluate the question-answering capabilities of embodied AI agents in complex environments, assisting researchers in exploring the integration of visual perception and language understanding.




