VCAS-Motion

VCAS-Motion数据集是由阿尔伯塔大学的研究团队创建，专注于自动驾驶中的视频类无关分割任务。该数据集包含520个序列，覆盖了8个不同的对象类别，提供了手动标注的密集掩码，以支持精确的实例分割。创建过程中，研究团队扩展了现有的KITTI和Cityscapes数据集，增加了标注序列和对象类别，以提高数据集的多样性和实用性。VCAS-Motion数据集主要应用于自动驾驶系统中的障碍物识别，尤其是那些在训练时未知的对象，旨在提高自动驾驶的安全性和鲁棒性。

The VCAS-Motion dataset was created by a research team at the University of Alberta, focusing on video-based out-of-distribution segmentation tasks in autonomous driving. This dataset contains 520 sequences covering 8 distinct object categories, and provides manually annotated dense masks to support accurate instance segmentation. During its development, the research team expanded the existing KITTI and Cityscapes datasets by adding annotated sequences and object categories, thereby improving the dataset's diversity and practicality. The VCAS-Motion dataset is mainly applied to obstacle recognition in autonomous driving systems, especially for objects unseen during model training, aiming to enhance the safety and robustness of autonomous driving.