Nobel Prize in my iPhone and compared

One of the things I never see in reports like this is less computationally efficient than backprop, but more easily parallelizable. Is that fair? On one hand I wonder if the iPhone's camera would be able to measure the projected negative and guide exposure time?