A breakthrough in the industry: NVIDIA taught the AI to handle the game graphics

NVIDIA for a long time is a real “engine for progress” in development of graphic solutions. Its graphics are still some of the best solutions on the market. But to infinity “to increase the gigabytes of VRAM” will not work and the industry needs fresh ideas. And that it is willing to give NVIDIA the world with his new technology that makes processing 3D graphics can not be implemented by the graphics processor, and advanced artificial intelligence system.

As the press service of the company NVIDIA have created the AI-algorithm, which creates virtual objects based on real movies, but unlike most similar solutions it does not work with separate frames, but with the entire stream at once. According to the company

“AI will allow developers and artists to create new interactive three-dimensional virtual worlds for automotive, gaming industry and for projects in virtual reality, just learning to live from the real world. It will reduce the cost and development time. Neural networks, and in particular generative models, change the way you create graphics.”

For technology demonstration, the research team developed a simple racing game powered by Unreal Engine 4, which allows players to navigate through an interactive environment generated by artificial intelligence. The neural network operates with high-level descriptions of scenes, such as a map segmentation that describe the characteristics of objects and where they are. Based on this, the AI “knows” whether a certain part of the image is a car or a building, as well as their boundaries and properties. Then the network imbues the scene with details that are learned from the source data and real videos.

For learning AI developers used NVIDIA Tesla V100 algorithm deep learning cuDNN, as well as thousands of videos of urban landscapes for a learning system. In addition to generating the virtual worlds, the AI can also create and animate 3D models of people.

“The ability to simulate and to reconstruct the dynamics of our world in the virtual space is important. In addition to purely scientific research, the training AI is the process of generating a continuous visual scale has a wide range of applications in robotics, computer graphics and so on.”