Nvidia Trained An AI To Impeccably Decrease Noise And Remove Watermarks From Pictures

A pretty picture can be created by artificial intelligence while potentially violating the law. The most recent AI technology of Nvidia can automatically mend the grainy photos. And this “mending” consists of eliminating the watermarks and text.

Nvidia, and a research team from MIT & Aalto University, taught the neural network of the AI by providing it a number of pictures. However, rather than providing it before-and-after photographs with corrupted as well as optimal models, the research team only allowed the AI to examine the corrupted images.

In their paper, the researchers said, “It is likely to study to reinstate signals without ever viewing clean ones, at times performance surpassing training utilizing clean examples. The neural network is equivalent to high-tech techniques that use clean exemplars—using accurately the same coaching method, and mostly without considerable problems in performance or training time.”

Practical uses of the AI consist of clearing out long exposure images s of the night sky captured by telescopes, as the astrophotography cameras often produce noise that can be flawed for stars. Also, the AI can be useful for medical purposes such as MRI that needs substantial post-processing to eliminate noise from pictures that are produced, so that physicians have a clean photo of what is happening in someone’s body. The AI of Nvidia can slash that processing period down considerably, which in turn decreases the time required for a diagnosis of a severe condition.

The team will be presenting their effort at the International Conference on Machine Learning to be held in Stockholm, Sweden.

Earlier, using a diverse method, Nvidia taught a deep learning system to change standard video into a slow motion, inserting frames subsequent to the capturing the video. By showing it numerous reference videos in the preferred slow-motion, the team trained the AI to determine how the missing frames were thought to appear.