Custom dataset for machine learning. (outdoor computer vision)
We have developed one of the largest domain-randomized simulators in the world for generating outdoor computer vision datasets.
Using openstreetmaps, and using thousands of 3d assets, we generate randomized, realistic recreations of real world scenes and neighbourhoods.
Using our proprietary Cycle-GAN models we can further enhance these rendered images.
Models trained on these datasets work flawlessly on real life images, because of the diversity and quality of our simulator and Cycle-GAN.