Honda Research Institute 3D Dataset (H3D)
收藏资源简介:
H3D数据集是由本田研究所创建的大型全环绕3D多目标检测和跟踪数据集,专门用于解决拥挤城市场景中的交通场景理解问题。该数据集包含160个拥挤和高度互动的交通场景,总计1,071,302个标记实例,分布在27,721帧中。数据集通过3D LiDAR扫描器收集,具有独特的数据大小、丰富的注释和复杂的场景,旨在推动全环绕3D多目标检测和跟踪的研究。为了有效且高效地标注大规模3D点云数据集,研究人员提出了一种标注方法,加速了整体标注周期。此外,H3D数据集还建立了一个标准化的基准,用于评估全环绕3D多目标检测和跟踪算法,支持未来算法的发展。
The H3D dataset is a large-scale surround-view 3D multi-object detection and tracking dataset created by the Honda Research Institute, specifically designed to address traffic scene understanding challenges in crowded urban scenarios. It contains 160 crowded and highly interactive traffic scenarios, with a total of 1,071,302 labeled instances distributed across 27,721 frames. Collected via 3D LiDAR scanners, this dataset features unique scale, rich annotations and complex scenarios, aiming to advance research in surround-view 3D multi-object detection and tracking. To efficiently and effectively annotate large-scale 3D point cloud datasets, researchers proposed an annotation method that accelerated the overall annotation cycle. Additionally, the H3D dataset establishes a standardized benchmark for evaluating surround-view 3D multi-object detection and tracking algorithms, supporting the development of future algorithms.




