Autonomous Intelligent Driving Advances with the Mapillary Vistas Dataset
Autonomous Intelligent Driving (AID) is a Munich-based startup that develops the full software stack for autonomous driving—from AI and Machine Learning for perception and prediction to localization, trajectory planning and interface to sensors and computers. AID is a subsidiary of AUDI AG and plays a central role in developing urban autonomous driving in the Volkswagen Group, providing technology that will be used across the group.
The Mapillary Vistas Dataset includes 25,000 images that have been manually annotated into 100 different object classes from many different locations (covering six continents) as well as different scenarios (time of day, weather, season). It provides training data for developing autonomous driving so that the algorithms used by self-driving cars can be properly trained to recognize different objects that occur in traffic situations, such as vehicles, pedestrians, lanes with different markings, traffic lights, etc.
AID’s mission is to enable mobility providers and car OEMs to drive fully autonomously in urban environments and beyond. Autonomous vehicles will increase mobility for people whose transportation today often depends on others, such as children and the elderly. They help reduce traffic congestion and the time people spend in traffic jams, making that time more productive, and create a safer traffic environment altogether, with fewer accidents on our roads and streets. Self-driving cars are at the forefront of future personal mobility and will have huge social impact, which is what guides the work of AID.
AID uses the Vistas Dataset as a high-quality and versatile resource for improving their Deep Neural Networks. Since it contains images from all over the world and from different kinds of cameras, it is ideal for improving the robustness of the models that AID is developing, and for verifying their performance in applications on potential markets all over the world. A clear benefit of the Mapillary Vistas Dataset, according to AID, is that the annotations are extraordinarily detailed which makes it very easy to map to it to their internal datasets.
"We are excited about Mapillary’s further development—their crowdsourcing-based data collection approach has the potential to become a major accelerator for autonomous vehicle development, and their AI technology can become a game changer for mapping as we know it," says Felix Friedmann, Tech Lead Perception of AID.
We look forward to seeing the results that AID will achieve with help of the Vistas and hope for further cooperation in the future. If you’re interested to hear more about licensing the Mapillary Vistas Dataset, please contact us.