Artificial Intelligence driven image uprez has been improving dramatically over the last few years. There are already commercial products like Topaz’ Video Enhance AI that seem like they will inevitably lead to the “computer enhance” featured in so many science fictions shows, where it is used to pull some tiny detail out of an image, or a license plate, etc.
Ironically, that’s probably one thing it cannot be used for.
Google have showcased a new approach whose development had been sidestepped in 2015 to allow other approaches to uprezzing to be explored. It seems like the original idea developed to fruition can create amazing results. From a Cined.com article:
How does that work? Well, the team took a high-resolution image and added noise until only pure noise remained. They then trained a neural network to reverse the process in order to recover the initial image. Finally, this trained neural network is used on images at 64×64 pixels to create superscaled versions. Jonathan and Chitwan then used this process in a stack to further improve the process. By stacking a 64×64 ? 256×256 model with a 256×256 ? 1024×1024 model, they were able to achieve impressive upscaling results.https://www.cined.com/new-google-ai-image-upscaling-makes-science-fiction-a-reality/
Making super resolution out of noise seems like alchemy to me. It will most certainly have many uses, particularly incorporating legacy material into more modern productions shot at higher resolutions.
As for the “computer enhance” problem. The “detail” is being invented, based on what the AI thinks it should look like. For realistic images this is excellent. For enhanced detail you could rely on for detective or legal purposes, almost certainly not. The detail is, quite literally, made up!