It turned out very plausible, but you can't play it yet.
A little more, and you will no longer have to leave the house – the dream of solipsists!
Developers from Intel have turned footage from the GTA V computer game into reality. As computer scientists say on their website, they have created a neural network that translates pixels into landscapes. To do this, the guys had to leave the stuffy office, get behind the wheel of a car and shoot terabytes of video, and then teach the neural network of life.
So far, the algorithm works at a low speed, but the developers hope that in the future they will begin to use it as a module in computer games, which will increase their realism.
If you are interested not only in the video, but also in how it all turned out, then here are some details. The authors decided not to create a new car simulator, but used GTA V, where most of the gameplay takes place behind the wheel. In addition to the frames from the game, they used the contents of the G-buffer – an additional channel of information that can be written into the final picture in order to further scare over it as you like. Rendering a 3D scene is split into two steps. In the first, the engine calculates geometric data and fills the G-buffer with frames that reflect the properties of objects. On the second, the engine uses this intermediate data on the properties of objects to render the lighting in the scene. Nothing is clear, but interesting. And it looks great.
The article uses materials from Enhancing Photorealism Enhancement.
Related materials:
- 6 reasons to play or not play Far Cry 6
- Japanese videographer recreates GTA on real streets, and the footage is difficult to distinguish from the game (video)
- GTA near Perm. The police are chasing a rowdy in shorts (VIDEO)
- The best examples of cynical ads in GTA V
- Pakistani politician praised the pilot of a passenger liner for an extreme landing, but it turned out to be footage from a GTA game
