We could soon be bidding goodbye to grainy photos. Nvidia and researchers from Finland’s Aalto University and MIT teamed up to use Artificial Intelligence (AI) to fix pixelated and grainy photos.
It is not the first time AI is being used to fix bad photos. But in previous iterations, the AI needed to look at before and after versions of the photos to be able to fix it. But this time, the AI doesn’t need to see a clean version of the target photo.
The team of researchers showed the AI two defective versions of a photo, and then trained it to remove pixels, additive Gaussian, Poisson and binomial noise. After studying over 50,000 samples, the AI is now able to clean up a photo better than many professional photo restorers can manage. What’s particularly interesting is how the AI is able to fix a photo is milliseconds.
Nvidia will be presenting this AI’s capabilities at the International Conference on Machine Learning in Stockholm. It says that there are a range of practical uses for this method. Astronomers clicking long exposure photos of the the space will benefit from noise-free results. In medicine, this method could clean up MRI scans in milliseconds, which will in turn help doctors to diagnose a serious condition.
Besides fixing noise, the AI is also able to remove visual artifacts like texts or watermarks from a photo. While impressive, it no doubt throws up privacy concerns. Photographers and companies have long used watermarks to prevent the misuse of their photos on the internet. It will be interesting to see how they react to an AI that can easily remove such watermarks in milliseconds.