You might not see most objects in near-total darkness, but AI can. MIT scientists have developed a technique that uses a deep neural network to spot objects in extremely low light. The team trained the network to look for transparent patterns in dark images by feeding it 10,000 purposefully dark, grainy and out-of-focus pictures as well as the patterns those pictures are supposed to represent. The strategy not only gave the neural network an idea of what to expect, but highlighted hidden transparent objects by producing ripples in what little light was present.
The researchers countered the blurring by giving it a physics lesson -- it knew how a defocused camera could produce blurring effects.
The result was an AI-based system that could find and reproduce hidden objects in lighting conditions would make even a dark room seem bright. That could help for nighttime photography, but the MIT research group is most interested in using the technology for medicine. It's normally hard to capture minute details in biological material without flooding them with light and radiation, risking damage to those tissues. This could avoid the problem by imaging tissues at much lower (and thus safer) light levels. It could help for astronomy, too, by detecting very faint objects without illumination.