NvidiaDemonstratedUse AI instead of traditional modeling or graphics rendering engines to draw virtual worlds in real time. The new approach takes advantage of conditional neural networks to analyze existing video and then apply 3D elements to the new environment. Nvidia claims that this technology greatly simplifies the creation of the 3D virtual world, because video-trained AI automatically renders buildings, cars, trees and other elements into the new 3D world without the need for painful scene element modeling. As shown in the figure, the video generated by AI is not too clear, and its advantage is that it is generated in real time.
The researchers released avideo(YouTube) demonstrates AI video rendering and scene replacement (replace the artist's own portrait with the characters in the original video).