This is all a bit tangential to PC gaming, but Nvidia researchers have been busy showing off the capabilities of its AI Deep Learning systems, demonstrating how its AI can now take footage filmed in winter and transform it into summer, or take a video of a sunny day and make it appear as if it’s raining.
Nvidia’s goal with Deep Learning is pretty clear - advance AI in order to develop additional tools and technologies which can tackle issues such as healthcare and autonomous vehicles. A large part of Deep Learning is focused on Neural Information Processing Systems (NIPS), the process by which an AI can take an image and understand what it is. As an example, it can identify not only what a bike is, but also understand where its pedals are, how old it looks, or whether it’s got flat tires.
Nvidia’s latest utilisation of NIPS is a natural extension of this. It’s an “Unsupervised Image-to-Image Translation Network”. The machine can have images and footage pumped in while the AI can identify it, manipulate it, and render out an edited version. In essence, the AI is learning imagination. It’s taking a winter image and imagining what it would look like in summer.
So to be clear - the AI has no prior knowledge whatsoever of what either of these two roads looks like during summer, or while it's raining. It understands what summer and rain are supposed to look like and it’s applying this to the winter and sunny images. The results of pretty astonishing. There are obviously a few errors, but for the most part, it manages to accurately put leaves on the trees, remove the snow, and even add the lines to the side of the road.
Quite how this would ever have any use in gaming I don’t know, but I thought it was a pretty cool indicator of just how Nvidia is utilising the same fundamental technology and hardware advancements as you’re using in your PC to run Assassin’s Creed Origins, to do all these incredible feats with AI.