NVIDIA, Aalto University, and MIT researchers have created an AI that can remove and automatically improve noise from grainy pictures. In several real-world circumstances, this technology can be useful in obtaining clear picture information such as MRI scans, astronomical imaging, and more.
“Recent profound field learning work has focused on training a neural network to restore images by showing example pairs of noisy and clean images,” writes NVIDIA. “Then the AI learns how to make the difference. This method differs because with noise or grain it requires only two input images.
Existing noise-reduction AI systems involve noisy as well as smooth input pictures, but the NVIDIA AI can restore pictures without showing what the noise-free picture looks like. Simply look at corrupted images examples.
The scientists trained the AI on 50,000 pictures and were able to generate impressive outcomes from the deep-learning algorithm.
See them below:
NVIDIA suggests that for this AI there are several real-world applications, as sometimes there are merely no “clean” examples to train with AI. Examples include low-light photography in some areas (e.g. astrophotography) where all we have is low-quality pictures.
Here’s a video by NVIDIA, Shows in AI in Action: Watch it