Aptiv plc has created an open-source dataset that developers of autonomous driving systems can access for free.
Dubbed nuScenes, the database consists of 1,000 20-second videos of complex driving scenarios recorded in Boston and Singapore. The archive is composed of 1.4 million images and 390,000 lidar sweeps, which Aptiv claims is the largest such collection in the world.
The scenes include 3D bounding boxes over key images—such as vehicles, pedestrians and animals—grouped in 23 classifications and eight attributes. A new feature also helps streamline object detection based on classification, localization, size, orientation, velocity and attribute estimation.
Aptiv says nuScenes detects, tracks and classifies various traffic situations, driving maneuvers and unexpected behaviors. Incorporating data from Boston and Singapore, allows users to design systems across different locations, weather conditions, vehicle types, vegetation, road markings and traffic for both left- and right-side-driving markets.
More than 1,000 users and 200 academic institutions have registered to access the nuScenes dataset. More information is available on Aptiv’s website.
Aptiv was formed in late 2017 when the business was spun off from Delphi Automotive plc. NuScenes was developed by the company’s Aptiv Autonomous Mobility (formerly nuTonomy) unit.
From this web page: