When Google’s Pixel 2 range rolled out last autumn, the smartphones were known to demolish absolutely every rival when it came to camera performance. While the competition was relying on a dual camera setup for the new portrait mode, the Pixel 2 managed to do it with a single camera and the power of machine learning, making portraits out of any still photo captured through its eyes. In fact, the Pixel 2’s camera app was so tempting to techies that they ported it out for older devices. While the port works only for some, Google wants to make sure that all Android manufacturers get to offer a portrait mode with a single camera sensor.
Google’s Research team has recently released their Semantic Image Stabilisation technology-based DeepLab-v3+ as an open source code project available for developers. The source-code can be used for any smartphone camera and Google wants third-party developers to improve upon the technology by implementing newer applications in various usage scenarios. DeepLab-v3+ utilises a neural network to identify certain elements in a photo such as a road, sky, person and dog. The semantic image segmentation technology puts a semantic tag to these elements to help the algorithms in applying depth-of-field applications.
While Google has made the source code available for use by anyone, it’s still required by third-party manufacturers to work on the image sensors. The DeepLab-v3+ will only bring the portrait mode to more smartphones while the image quality is still dependent on how an individual manufacturer optimises the software and hardware.
