OpenScan
收藏资源简介:
OpenScan数据集由香港城市大学和华南理工大学创建,专注于广义开放词汇的3D场景理解,涵盖了八个代表性的语言方面,如功能、属性、材料等。该数据集基于ScanNet200构建,包含超过15万条属性注释,旨在评估模型对抽象对象属性的理解能力。创建过程中,结合了知识图谱和人工注释,确保数据质量。OpenScan的应用领域包括自动驾驶和机器人技术,旨在解决3D场景中对象属性的识别和理解问题。
The OpenScan dataset, developed by City University of Hong Kong and South China University of Technology, focuses on generalized open-vocabulary 3D scene understanding. It covers eight representative linguistic aspects, including function, attribute, material and others. Built upon the ScanNet200 dataset, OpenScan contains over 150,000 attribute annotations, aiming to evaluate models' ability to understand abstract object attributes. During its creation, knowledge graphs and manual annotations were combined to ensure data quality. The application fields of OpenScan include autonomous driving and robotics, with the goal of solving the problems of object attribute recognition and understanding in 3D scenes.




