Artificial intelligence not only continues to master creative professions but also threatens to compete with game developers. At a recent conference, NVIDIA experts presented a neural network capable of creating virtual interactive worlds based on a real environment.

NVIDIA will entrust the creation of virtual neural network worlds

For the training of the algorithm, materials were used, shot in several cities by unmanned vehicles cameras. Then he was given the opportunity to independently recreate «what he saw» with the help of the Unreal Engine 4 graphics engine used in the development of many modern hits. As a result, the computer managed to create an uncomplicated auto simulator based on the recording obtained from the DVR.

When creating a virtual environment, developers are often faced with the high cost of production of the next blockbuster — because each object must first be drawn manually. Elements created by artificial intelligence can be modified and modified in the traditional resource editor, reducing the time to create game locations.

NVIDIA will entrust the creation of virtual neural network worlds

Such an approach will significantly simplify the development process by automating it and partially placing the routine duties of designers on a neural network that is continuously being trained in the process of studying the real world.

“NVIDIA has been inventing new ways to create interactive graphics for 25 years, and for the first time, we can do it with a neural network. Neural networks will change the way graphics are created. This will allow developers to create new scenes much cheaper, «said Brian Catanzaro, vice president of applied deep learning at NVIDIA.

In the demo video, the developers told about other possibilities of the algorithm. For example, one of its authors created a three-dimensional model of his colleague dancing the popular dance Gangnam Style. The expert noted that the virtual copy of his partner dances much better than a real programmer, who served as a prototype for AI, could do it.

The timing of the commercial launch of the neural network is not yet known.