One of the problems with modern game development is the need to improve the texture under increasing resolution. However, because of this the games no longer cope with even the most powerful graphics card.
To solve the division Xbox experimenting with machine learning technologies and artificial intelligence to create a system to scale textures in real-time. This will reduce the load on the video card and use the same textures for different resolutions.
While the project is at an early stage and, in fact, not out of the experimental stage. The developers managed to improve texture in one game, having trained the AI on a set of examples. But doing so for all the games too long and expensive. It is planned to create a universal algorithm for this. And processing will take cloud service Azure.
945ca4ffdb933d34_848xH.jpg
Technically this is similar to the results of the system ESRGAN (Enhanced Super-Resolution Generative Adversarial Networks), which have improved the textures of the original Doom, Half-Life and other past projects.
It seems that the company has been working in this direction. Previously, she has released DirectML that allows “on the fly” to increase the resolution from FullHD to 4K. Is NVIDIA Deep Learning Super-Sampling (DLSS), which uses the tensor kernel in GPUs Turing for improved game graphics.
It is not excluded that in the foreseeable future will see the same technology that allows you to optimize textures with AI in real time.