4K Graphics On Cheap Hardware With Neural Networks?

Original article was published on Artificial Intelligence on Medium


4K Graphics On Cheap Hardware With Neural Networks?

Overview of the paper “Neural Supersampling for Real-time Rendering” by Xiao et. al.

For VR gaming to truly take-off, we have to make the experience much more life-like and realistic. This requires rendering the graphics at very high-refresh rates and at least in 4k resolution so that the virtual world looks and feels like the real world. However, in order to do so, we currently require extremely high end and costly hardware, which is why VR gaming has not become very widespread and popular.

So today, I want to share a recent paper from Facebook reality labs that looks to address this problem with the help of neural networks. The paper is titled “Neural Super-Sampling for Real-Time Rendering” by Xiao et. al. and it introduces a deep learning method to up-sample the image rendered by a game engine in real-time allowing us to maintain the high refresh rate as well.