Generating Virtual Environments from Short Videos

Creating 3D Models from Flat 2D Images
March 10, 2020
Ambient Computing Expands
March 10, 2020

Generating Virtual Environments from Short Videos

Chip designer Nvidia is teaching A.I. to build realistic 3D environments from short video clips.

Chip designer Nvidia is teaching A.I. to build realistic 3D environments from short video clips.

The method builds on previous research on generative adversarial networks (GANs). Nvidia’s system generated graphics taken from open-source data sets used by the autonomous driving field. Using short clips segmented into various categories (buildings, sky, vehicles, signs, trees, people) the GAN was trained to generate new, different versions of these objects.

Future applications of automatically-generated virtual environments are vast: think training environments for logistics (warehouses, factories, shipping centers), urban planning simulations, even testing customer flow scenarios within amusement parks and shopping centers.