While we’ve seen loads of examples of AI making games look better over the last couple of years, I’ve never seen anything like this approach, that’s able to take Grand Theft Auto V and through the power of magic make it look so damn real.
This is a project by Stephan R. Richter, Hassan Abu AlHaija and Vladlen Koltun at Cornell University, culminating in a paper called Enhancing Photorealism Enhancement. Both the paper and the accompanying video get pretty heavy on technical details, so here’s the basic summary of what they’re doing:
We present an approach to enhancing the realism of synthetic images. The images are enhanced by a convolutional network that leverages intermediate representations produced by conventional rendering pipelines. The network is trained via a novel adversarial objective, which provides strong supervision at multiple perceptual levels. We analyze scene layout distributions in commonly used datasets and find that they differ in important ways. We hypothesize that this is one of the causes of strong artifacts that can be observed in the results of many prior methods. To address this we propose a new strategy for sampling image patches during training. We also introduce multiple architectural improvements in the deep network modules used for photorealism enhancement. We confirm the benefits of our contributions in controlled experiments and report substantial gains in stability and realism in comparison to recent image-to-image translation methods and a variety of other baselines.
In short: machines can only work with what they’re told, so by improving the quality of what’s being fed to them, and the way they’re being fed, the results can be drastically improved. Like this:
If you want to get really into it, even more so than in the video above, you can read the whole paper here.