Computational photography is a big thing these days. Not only when it comes to smartphones but also with software for the desktop using photos shot with a “real camera”. One particular area where it’s beneficial is noise reduction. It’s built into many smartphone camera apps automatically and it’s available for the desktop in applications like Noise Ninja and Topaz DeNoise AI, not to mention denoising features built into apps like Lightroom. Google’s new Neural Radiance Fields (NeRF) AI, titled RawNeRF (AKA NeRF in the Dark), however, blows them all away, with images that are not only so good they look almost flawless but it can then go on to generate completely artificial 3D scenes that you can move around in after the fact, as well as adjust exposure, gamma & tone mapping and even focus – including bokeh balls!
Presented by Two Minute Papers, the new Google AI shown off in the video is probably the best noise reduction tool I’ve seen to date, for any system, even if we take away all of the other things it can do, like 3D reconstruction and creating bokeh that would make an f/1.2 lens jealous.
With smartphones getting more and more powerful with what seems like each passing month, we might not have to wait long to see this level of denoising technology come to the cameras in our pockets. And if you want to try it on the desktop, you can do so already. All of the code is available to download from GitHub and try for yourself and there’s more information about the process here. The possibilities of this technology, even with the level it’s already at are pretty mindblowing. It offers a lot of potential for everything from smartphone cameras all the way up to TV shows and Hollywood movies. Hopefully, within the next year or two, this (or something like it) will start being implemented in a lot of the editing tools we use daily!
