UT Campus Object Dataset (CODa)
收藏资源简介:
UT Campus Object Dataset (CODa)是由德克萨斯大学奥斯汀分校创建的一个移动机器人自我中心感知数据集,包含8.5小时的多种传感器数据,如同步的3D点云和立体RGB视频。数据集提供了58分钟的地面实况标注,包含130万个3D边界框,用于53个语义类别,以及5000帧的3D语义标注,用于城市地形。数据集通过在室内外环境中重复遍历相同的地理位置,涵盖了广泛的天气条件和一天中的不同时间。CODa旨在用于城市环境中自主导航的3D感知和规划研究,通过实验证明,使用CODa训练的3D物体检测性能显著高于现有数据集,即使在采用最先进的领域适应方法时也是如此。
UT Campus Object Dataset (CODa) is an ego-centric perception dataset for mobile robotics developed by The University of Texas at Austin. It encompasses 8.5 hours of multi-sensor data, featuring synchronized 3D point clouds and stereo RGB videos. The dataset offers 58 minutes of ground truth annotations, comprising 1.3 million 3D bounding boxes spanning 53 semantic categories, alongside 5000 frames of 3D semantic annotations for urban terrain. By repeatedly traversing identical geographic locations in both indoor and outdoor settings, the dataset covers a diverse range of weather conditions and varying times of day. CODa is targeted at research on 3D perception and planning for autonomous navigation in urban environments. Experimental validations have shown that the 3D object detection performance trained with CODa is significantly higher than that of existing datasets, even when utilizing state-of-the-art domain adaptation methods.




