Scale AI is a start-up specializing in databases used to train artificial intelligence systems. The young shoot combines software and human supervision to process and label images from Lidar sensors and data from 3D mapping. Scale AI announced on May 20, 2020, in collaboration with sensor manufacturer Hesai, that it will make publicly available (open source) PandaSet, a set of data sets for autonomous driving.
Fusion of sensors is possible
These data are free for commercial or academic use. The platform brings together more than 48,000 images from cameras and 16,000 Lidar scans, which corresponds to more than 100 eight-second scenes, Scale AI list. The data has 28 different annotation classes for each scene and 37 semantic labels for the majority of them. These data were taken in real situations, in urban areas where vehicles are confronted with other road users and where the lighting can vary depending on the day.
PandaSet also includes Scale AI’s sensor fusion technology. Teams using this platform can merge images from different sensors to obtain a single point cloud.
Autonomous vehicle tests stopped
Scale AI hopes to meet a need. The Covid-19 pandemic has slowed down, or even stopped testing of autonomous vehicles. Many companies no longer collect data from their test vehicles and are forced to turn to track testing or digital simulation. However, they need quality, well-labeled data to train their algorithms. Scale AI therefore hopes to seduce these players with its ready-to-use datasets from real situations. Note that outside the autonomous driving sector, Scale AI counts Airbnb, Pinterest or OpenAI among its customers.