ApolloScape Autonomous Driving Dataset

ApolloScape is part of the Apollo autonomous driving project, which aims to promote innovation in all areas of autonomous driving, from perception, navigation to control. The dataset is still being updated in terms of data size, label density, and tasks.

The dataset includes (currently):

• Dense semantic 3D point cloud of the environment (20+ driving sites)
• Driving videos in stereo (100+ hours)
• High-precision 6-DOF camera orientation (translation ≤ 50mm, rotation ≤ 0.015°)
• The time the video is on the same website (morning, noon, evening)
• Dense per-pixel per-frame semantic labeling (35 categories, 144K+ images)
• Per-pixel landmarks (35 categories, 160K+ images)
• Semantic 2D Instance Segmentation (8 categories, 90K+ images)
• 2D car keypoints and 3D car instance labels (70K cars)

At the same time, the dataset is divided into the following subsets according to autonomous driving related tasks:

• Scene Parsing
• 3D Car Instance
• Lane Segmentation
• Self Localization
• Trajectory
• 3D Lidar Object Detection and Tracking
• Stereo
• Inpainting