If TensorFlow, PDAF pixels, and semantic segmentation sound like your kind of jam, check out this deep dive into mobile imaging from Google research lead Marc Levoy. He goes into some detail about how the team behind the new Pixel 2 trains neural network, detects depth, and synthesizes pleasing, realistic bokeh even with a single-lens device. [Update: There’s a higher-level, less technical version of the post if you’d prefer.]