Google makes AI ‘portrait mode’ tech open source and opens up Maps API to gaming devs
GOOGLE HAS ANNOUNCED plans to open up data from several parts of its empire to allow third parties to make use of it.
Firstly, third-party Android users will be able to take advantage of the AI technology that powers portrait mode in the Pixel 2 and Pixel 2 XL .
The open source version, called DeepLab-v3+, is not identical to that within Google’s flagship phone, but Google says it will yield similar results.
The main upshot of this technology will be to create a feeling of depth in photographs without the need for a dual lens camera. As Google tries to shave costs to empower the ‘next billion’ Android phones, there’s a lot of sense in this.
It does this using a neural network which has learned to identify shapes of humans from the background and focuses on them while simultaneously classifying them for future reference.
The new tech uses Google’s TensorFlow technology and is designed to be used server-side meaning the best results will be when you’re connected to the Internet.
The explanation gets very techie but if you’re into that kind of thing here is part of it: «Since the first incarnation of our DeepLab model [4] three years ago, improved CNN feature extractors, better object scale modelling, careful assimilation of contextual information, improved training procedures, and increasingly powerful hardware and software have led to improvements with DeepLab-v2 [5] and DeepLab-v3 [6].
«With DeepLab-v3+, we extend DeepLab-v3 by adding a simple yet effective decoder module to refine the segmentation results especially along object boundaries.»
Google is hoping that companies will be able to use the technology to improve cameras in existing and upcoming phones and maybe even come up with some ideas of their own about how to make the most of it.
At the moment there is no indication of if and when an event this will be implementing this and indeed Google is offering no promises that it will significantly improve photo quality just that it may well do.
In a separate move, the company has also launched an API for Google Maps which make it a lot easier for developers to incorporate real-world data into games.
This will mean the next Pokemon Go be a lot easier to build and an additional API bolt on for the Unity game engine will bring Google Maps an even wider range of games.
Three games are already in the pipeline but will use Google Maps data including spin-offs from Jurassic World, The Walking Dead, and Ghostbusters.
Google is on something of a roll right now in terms of making information available. Earlier this month it announced that it will be making its AI training programme for Googlers is available to anyone with a yearning to learn.
Yesterday, it released makers’ kit for anyone that wanted to try building its AI-powered musical instrument NSynth .
Plus of course, last week saw the announcement of the first developers build of the as yet unnamed Android P . µ
