NVIDIA has joined forces with MIT and Aalto University to create Noise2Noise, an AI that is able to remove the ‘noise’ from grainy photos.
The system uses a deep learning technique to make it work, a form of machine learning that teaches AI how to put together various forms of media such as images, video or text.
The previous way they tried to train the AI was by making it reconstruct photos by first feeding it complete photographs. The second method involves corrupted data – bad photographs – that the AI rebuilds.
The AI was given 50,000 noisy images which helped it learn what noise was in the first place. Noise2Noise removed it, bringing those pics closer to the original.
Sure, they are not as sharp as the source pictures and some details are missing, but the outcome is still considered a success.
This technology could be easily used in the medical field for MRI scans, for example, which could be rendered more clear.
For those of you who can’t wait to be able to take decent photos with your phones in low-light conditions, yes, the AI works even on digital cameras or smartphones.
For the time being though, it’s still a proof-of-concept, which means there will be some time before non-professional users, or anyone not working for NVIDIA or MIT, can get their hands on it.