ApolloScape

ApolloScape是由百度研究院创建的大型、多样化的自动驾驶数据集，包含丰富的标注信息，如全场景语义密集点云、立体视频、每像素语义标注、车道标线标注、实例分割、3D车辆实例等。数据集覆盖多个地点、城市和时间段，支持多种任务，如3D重建、自我定位、语义解析等。创建过程中，开发了多种工具和算法以加速标注过程，如联合3D-2D分割标注、视频中的主动标注等。数据集应用于多传感器融合和多任务学习领域，旨在推动计算机视觉技术的发展。

ApolloScape is a large-scale, diverse autonomous driving dataset created by Baidu Research. It comprises a rich set of annotated data, including full-scene dense semantic point clouds, stereo videos, per-pixel semantic annotations, lane marking annotations, instance segmentation, 3D vehicle instances, and more. The dataset covers multiple locations, cities and time periods, supporting a wide range of tasks such as 3D reconstruction, ego-localization, semantic parsing and so on. During its development, various tools and algorithms were developed to accelerate the annotation process, such as joint 3D-2D segmentation annotation and active annotation in videos. It is applied in the fields of multi-sensor fusion and multi-task learning, aiming to promote the advancement of computer vision technology.