The Parallel Domain data generation platform offers procedurally generated worlds with which datasets can be generated. These worlds contain a variety of road networks and other elements that make them effective in training computer vision models.
Key concepts for understanding our worlds are introduced below:
Map: A collection of road networks, junctions, building footprints, and other semi-abstract data which describes a world. Maps are both collected and generated internally by Parallel Domain for use in our world generation pipeline, or they are provided by customers in common HD map formats which are then converted to Parallel Domain's internal map format for use in our pipeline.
World: A world is a collection of 3D geometry representing roads, terrain, buildings, and more. Each world has an accompanying map file that enables the procedural generation of scenarios that output synthetic datasets. As a part of the scenario creation process, dynamic agents and other props are placed around the world, and conditions such as lighting and weather can be altered.
Region: A region is a collection of parameters that apply a style or look to a world. These parameters are used to enable/disable/modify identifying characteristics of a particular place in the real world, such as the types of vegetation that appear on the side of a highway or the architectural styles present in an urban scene. For example, a map may define a 4-way intersection that is present in the world seen in dataset renders, but the region applied to that world may alter the road lane width, the road surface and wear condition, and the profile of the curbs.
We create a variety of regions for our dataset generation pipeline (e.g., US Highways, Japanese Highways, Tokyo). Browse the sections below to get a sampling of the regions you can access. We are happy to discuss specific geographies or regions you might be interested in, please feel free and reach out to our sales team to learn more.