Companies developing autopilot algorithms for cars are usually forced to independently collect data to train the system. To do this, it is desirable to have a fairly large fleet of vehicles operating in heterogeneous conditions. As a result, development teams that want to put their efforts in this direction are often unable to do so. But recently, many companies developing autonomous driving systems have begun publishing their data to the research community.
One of the leading companies in this field, Waymo, owned by Alphabet, followed a similar path and provided researchers with a set of data from cameras and sensors collected by its fleet of autonomous vehicles. The package contains 1000 road recordings of 20 seconds of continuous motion, captured at 10 frames per second using lidars, cameras and radars. The objects in these recordings are carefully labeled and have a total of 12 million 3D labels and 1,2 million 2D labels.
The data was collected by Waymo machines in four American cities: San Francisco, Mountain View, Phoenix and Kirkland. This material will be an important aid for programmers developing their own models for tracking and predicting the behavior of road users: from drivers to pedestrians and cyclists.
During a briefing with reporters, Waymo research director Drago Anguelov said, βCreating a dataset like this is a lot of work. It took many months to label them to ensure that all significant parts met the highest standards that can be expected, confident that researchers have the right materials to help make progress.β
In March, Aptiv became one of the first major self-driving vehicle operators to publicly release a dataset from its sensors. Uber and Cruise, the autonomous division of General Motors, also presented their materials for the development of autopilot to the public. In June, at the Computer Vision and Pattern Recognition conference in Long Beach, Waymo and Argo AI said they would eventually release datasets. Now Waymo has delivered on its promise.
The company also claims that its data package is more detailed and detailed than those offered by other companies. Most previous sets were limited to just camera data. The Aptiv NuScenes dataset included lidar and radar data in addition to camera images. Waymo provided data from five lidars, compared to the only one in the Aptiv package.
Waymo also announced its intention to continue providing similar content in the future. Thanks to this kind of action, software development for traffic analysis and vehicle control can receive additional impetus and new directions. This will also help student projects.
Source: 3dnews.ru