Nvidia and scientists teach AI to remove graininess from photos

Scientists have developed an artificial intelligence system that can automatically remove noise, specks, and other distortions from pictures.

The technology, called Noise2Noise AI, was developed by researchers from Nvidia, Aalto University in Finland, and MIT. The researchers used 50,000 pictures, as well as MRI scans and computer-generated noisy images, to train the system. According to the research paper, the AI can remove enough noise to make images usable again without ever seeing a clean image.

Nvidia Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework were used to train up the system.

The technology has been trained to remove noise without needing to understand what a clean image looks like, which until now, such AI work has focused on training a neural network to restore images by showing example pairs of noisy and clean images.

"It is possible to learn to restore signals without ever observing clean ones, at performance sometimes exceeding training using clean exemplars," the researchers said.

They added that 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".

The system was tested using three different datasets to validate the neural network.

Scientists said there were several real-world situations where obtaining clean training data is difficult: low-light photography (e.g., astronomical imaging), physically-based image synthesis, and magnetic resonance imaging.

"Our proof-of-concept demonstrations point the way to significant potential benefits in these applications by removing the need for potentially strenuous collection of clean data," said the paper's authors. "Of course, there is no free lunch we cannot learn to pick up features that are not there in the input data but this applies equally to training with clean targets."