In critical domains such as national security, federal asset management, and economic monitoring, the use of computer vision to extract information for decision-making and analysis is becoming increasingly urgent.  The amount of video, photography, and other forms of sensor-based information being collected far outstrips our national capacity to staff and monitor data feeds with human eyes alone.

Unfortunately, computer vision accuracy and performance is limited by the datasets used to train algorithms. Data may be expensive to obtain, may not capture rare and unusual objects or scenarios, and, in some cases, may be extremely high risk or even impossible to collect.  Rapidly new applications of imagery such as SAR and x-ray stretch the capacity of humans to label and identify.

Physically accurate synthetic, or engineered, computer vision data offers the opportunity to overcome some of the challenges of using real sensor data. Synthetic data can be designed for specific sensor types or to have known distributions of objects or scenarios for algorithm training. Synthetic data is cheap and can be processed rapidly in the cloud and on demand.

In this webinar, Rendered.ai shared information about:

• • • • Opportunities for using synthetic data to supplement real sensor data collection, overcoming some of the challenges of using real data alone
      • Examples from a range of industries of using synthetic data for scenarios such as object detection in RGB and hyperspectral imagery, early experiments with synthetic SAR collection, and use of x-ray simulation for security inspection
      • Use of industry standard simulation tools such as NVIDIA Omniverse and RIT’s DIRSIG in the Rendered.ai platform as a service for generating massive datasets in the cloud