Who says AI, machine learning, and computer vision has to be cold and boring to anyone except computer scientists? There has been an increasing number of studies and experiments that try to employ the fruits of these advanced studies to more approachable, sometimes whimsical, applications. And what can be more whimsical than turning your perfect selfie into a caricature of yourself. And that is exactly what Google Allo is now capable of doing, all on its own, without your help or your artistic input.
Say you want to express yourself in a way only an emoji or sticker can. And you find that there is nothing on Allo that fits your intended expression. With this new feature, you can just make a face, take a selfie, sit back, and watch Allo do its magic.
That magic, of course, is thanks to neural networks and machine learning. But for this particular application, the conventional wisdom of computer vision isn’t at all helpful. Because of the way people take selfies from different angles and different lighting conditions, it would be nearly impossible for computer vision to simply map your selfie to some piece of art.
Instead, Google’s algorithm picks out unique features of a face that it abstracts to replaceable features like hairs, eyes, etc. There is also some human element to this whimsy, from the illustrations drawn by artists to the human “QA” that rate the combinations the neural network tries to make.
A feature that could provide lots of fun to users, this selfie-to-sticker feature is also lots of fun to the geeks at Google. The lessons learned and applied in this experiment can be used to create more efficient ways to recognize faces. Which, considering this is Google, might sound worrying in the long run. Either way, the feature is now rolling out to Android users, with iOS following soon.