Building safe autonomous vehicles demands extremely robust learning processes. Today’s training and testing of autonomous systems in real-world environments is very costly and potentially biased to predefined fixed scenarios. Unbiased synthetic scenarios enhances the ability to train systems for challenging cases rarely encountered in controlled real-world environments, with the additional benefit of much higher variability at a fraction of cost.
ANYVERSE™ is a Data as a Service solution capable of delivering massive machine-generated datasets in the range of millions of tagged images for both training and testing/validation processes at a much lower overall cost. The ANYVERSE production pipeline can be adapted and customized to the specific needs of the training teams.
ANYVERSE’s Variations Engine can produce unbiased variations of arbitrary digitally generated scenarios, lighting (traffic, street, buildings, sun position, night conditions) and scenery features such as atmospheric effects, object damages, other vehicles, road layout and pedestrians. Millions of virtual miles can be trained and tested in a fraction of the time, guaranteeing a competitive advantage over teams relying exclusively on real-world datasets.
ANYVERSE datasets can deliver semantic, boundary and instance segmentation masks. Dense annotations are also provided including categories like road, person, rider, sky, ground, terrain, car, bus, sidewalk, traffic light, traffic sign and much more. Because of its synthetic nature ANYVERSE datasets are always ground truth, there is no room for wrong annotations.
An evaluation synthetic dataset is freely available, including full size images and tagged metadata. A specific dataset focused on traffic lights recognition and classification is also available. These datasets can be valuable for teams working on either autonomous vehicle development or driver assist solutions.
Contact us to know more about the ANYVERSE solution and get access to the evaluation dataset.