Motion4D
收藏资源简介:
Motion4D是由华中科技大学与南洋理工大学联合创建的首个面向合成到真实运动预测研究的大规模合成4D LiDAR数据集。该数据集包含1,370个动态序列,总计124,000帧点云数据,涵盖了丰富多样的交通运动模式,其数据通过基于物理的仿真流程生成,利用CAD资产和射线投射算法渲染高保真序列。该数据集专为合成到真实域自适应场景设计,旨在解决自动驾驶领域中真实运动标注成本高昂的难题,为运动预测模型的训练与评估提供基准支持。
Motion4D is the first large-scale synthetic 4D LiDAR dataset dedicated to synthetic-to-real motion prediction research, jointly developed by Huazhong University of Science and Technology and Nanyang Technological University. This dataset includes 1,370 dynamic sequences with a total of 124,000 frames of point cloud data, covering a diverse range of traffic motion patterns. The data is generated via physics-based simulation pipelines, and high-fidelity sequences are rendered using CAD assets and ray-casting algorithms. Specifically designed for synthetic-to-real domain adaptation scenarios, this dataset aims to resolve the high-cost challenge of real-world motion annotation in autonomous driving, and provides benchmark support for the training and evaluation of motion prediction models.



