Training AI to see in the dark and take low-light photography to a new level

When shooting in low light, you need to either shoot at long exposures on a tripod or crank up the ISO if you want to shoot handheld. A group of researchers at the University of Illinois Urbana-Champaign and Intel are bringing the best of both worlds. They’ve trained AI to process low-light images so they’re much cleaner and more usable than grainy photos where ISO is too high. This tech could let you shoot at faster shutter speeds and lower ISO and still get with sharp, clear photos.

When captured at short exposure times, the photos would appear almost black. The long-exposure equivalents were then used to train the AI to recreate the photos of good quality and proper exposure.

In the video demonstration, you can see photos taken at different exposure times, ISO values and with different gear (Fuji X-T2, Sony a7s II and iPhone 6s). They were all taken at low-light conditions, and you can see the results after the AI reconstructed them. The results are pretty impressive: the noise is significantly lower, and the colors look better as well. Here are a couple of examples:

 

 

The researchers have provided the code on GitHub, and you can read more about the research in the paper they published. And if you want to see more example images, you can check out the gallery here.