A team of NVIDIA researchers, in partnership with researchers from Aalto University and Massachusetts Institute of Technology (MIT), has shared details of a new artificial intelligence (AI) program that can remove grain from images with such accuracy that it's almost scary.

'Using NVIDIA Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework, the team trained [its] system on 50,000 images in the ImageNet validation set,' says NVIDIA in its announcement blog post.

What's incredible about this particular AI is its ability to know what a clean image looks like without ever actually seeing a noise-free image. Rather than training the deep-learning network by giving it a noisy image and a clean image to learn how to make up the difference, NVIDIA's AI is trained using two images with different noise patterns.

'It is possible to learn to restore signals without ever observing clean ones, at performance sometimes exceeding training using clean exemplars,' say the researchers in a paper published on the findings. The paper goes so far as to say '[The neural network] is on par with state-of-the-art methods that make use of clean examples — using precisely the same training methodology, and often without appreciable drawbacks in training time or performance.'

In addition to being used on photographs, researchers note the AI will also be beneficial in scientific and medical fields. In particular, the researchers detail how magnetic imaging resonance (MRI) scans — which are very susceptible to noise — could be dramatically improved using the program, leading to improved diagnoses.

The team behind the AI will present their work at the International Conference on Machine Learning on July 12, 2018.