We’ve collected and annotated millions of video frames to train our neural networks. Our data collection is powered by a fleet of aircraft, including airplanes, helicopters, and drones. In total, through flights on our own platforms as well as those of partners and customers, we’ve gathered over 1,000 hours of video footage using our onboard camera systems across Europe, North America, and South America.
These recordings capture a wide range of landscapes and vegetation zones for visual navigation tasks. They also include tens of thousands of annotated traffic encounter scenarios and thousands of runways, all curated to train and validate neural networks.
But we don’t just focus on quantity—we prioritize diversity. To ensure our system can reliably detect and identify gliders, for instance, we equipped a glider with our technology and participated in glider competitions. Likewise, to gather balloon data, our system took part in ballooning events.
Our ambition is clear: to build and own the world’s most comprehensive proprietary dataset for machine learning in aviation. And we are delivering on that goal.
