The new AI from Nvidia is capable of significantly reducing the noise on a photo and is even capable of removing watermarks.
The AI can “map corrupted observations to clean signals… [sometimes] without ever observing clean ones.” The AI is capable of training itself using a good and a bad version of an image. However, what makes it special is the fact that it does not need (sometimes) a good image / good example to train itself. The idea is an expansion of the concept of deviation-minimizing estimators, or M-estimators, which tell you how to estimate true data from a set of unreliable data.
It can teach itself by looking only at noisy photographs. Of course, there are some specific mathematical caveats, but this can help astronomy in a big way!
Subscribe to our newsletter
Then there is the fact that it is also capable of deleting watermarks with ease, which is kinda scary. Nevertheless, it will most certainly help photographers when it comes to post-processing.