The Pixel 2's camera package is excellent for a lot of different reasons, and one of my favorite features is easily its portrait mode. The Pixel 2 continues to create the best-looking portrait-style shots even with phones like the Galaxy S9 and iPhone X out in the wild, and now Google's doing the unexpected by taking the system that makes it possible and opening it up to other developers.
DeepLab-v3+ is the technology in question, and this essentially gives each pixel its own label to identify objects in the foreground and background. DeepLab-v3+ is being added to Google's TensorFlow development platform, and as such, developers will be able to integrate this same framework into their apps.
Per Google's announcement post on its Research Blog:
Since the first incarnation of our DeepLab model three years ago, improved CNN feature extractors, better object scale modeling, careful assimilation of contextual information, improved training procedures, and increasingly powerful hardware and software have led to improvements with DeepLab-v2 and DeepLab-v3. With DeepLab-v3+, we extend DeepLab-v3 by adding a simple yet effective decoder module to refine the segmentation results especially along object boundaries.
DeepLab-v3+ isn't the sole item that powers portrait mode on the Pixel 2, but Google notes portrait mode is "an example of what this type of technology can enable." So, while Google isn't necessarily open-sourcing the Pixel 2's portrait mode, it is allowing developers to use the same system that made it possible.
