A group of researchers from the University of Illinois joined forces with a team from Intel to develop a neural network that is able to perform post-processing enhancements on extreme low-light photographs.
The AI is able to make photographs (that are otherwise pitch black or full of noise) look clean and colorful.
The researchers created a SID (See-in-the-dark) data set, a group of over 5,000 short exposure images in RAW format and passed it over to a deep learning system. The AI was trained to compare the information within the low light images to the information of photographs taken at longer exposure.
These results are groundbreaking, especially when you consider that photographers require tripods and long exposure shots to get the right shot.
The AI can, theoretically, be optimized to provide real-time image processing. In the future, this AI could be used to develop devices that would allow us to see in the dark perfectly.