At Facebook's day-two F8 keynote, Joaquin Quiñonero Candela, one of the company's AI honchos, just provided an update on the technologies the company is developing to analyze video. The company is able not just to identify multiple people in a clip but also what they're doing: sitting, standing, walking, waving.
It's also working hard to push AI onto the phone rather than relying on servers in the cloud. One project involves analyzing 2D video—most phones only have one lens on each side—and extrapolating 3D space from a flat image. That's useful for effects such as real-time video selfies that make you look crisp but pleasantly blur the background.