There may be no such thing as bad press, but there is such a thing as a bad API, as Microsoft learned past week with their dubiously viral how-old.net. Introduced by VP Joseph Sirosh at Microsoft’s Build conference, the technology offered a taste of computer vision and deep learning to consumers, allowing them to track their age through pictures of their faces. There’s just one problem- it’s not very good.
It’s not just that the experimental age detector is inaccurate, rather, that it’s completely arbitrary. Microsoft’s team seems to have the ability to detect faces down pat, but when a teenager’s furrowed brow is indicative of middle age, or when a father’s profile can appear significantly younger than his fresh-faced son’s… well, it might be time to go back to the drawing board.
What was supposed to be a glimpse of the future ended up exactly the opposite: a superior API utilizing this technology has been around for quite some time now, and it’s both more accurate and more complex than what Microsoft has on offer. We should know- our app is built on it!
The engine underneath PhotoTime’s hood is Orbeus’ ReKognition API, and there’s just no comparing the two. We’ve run a few of the same images through Microsoft’s API and through our own- you can judge for yourself which is more accurate.
And it’s not just accuracy, but breadth of content that ReKognition excels at. While Microsoft is stuck working out the kinks in detecting age, ReKognition has become powerfully accurate at discerning conceptual features of photos of all kinds. This is part of what makes PhotoTime such a valuable app: not only does it seek out the basic content of the photo, it also has an awareness of the context of the photo. Detecting locations, subtle emotions, even social situations: this was all a challenging, iterative process. But it’s not experimental tech, it’s right here for you to explore for yourself, and if you’ve got PhotoTime, it’s already making your life easier.
If you were one of the thousands and thousands of people disappointed by MS’ rudimentary attempts at computer vision, check out ReKognition and PhotoTime. There’s virtually no end to the applications of an API as slick as this, and far from needing the kinks worked out, ReKognition is only growing more powerful by the day. Trust us, after seeing how-old.net you’ll be amazed at how shallow our deep learning isn’t.
So give us some feedback: did we get your age right? How about your emotions? What else would you like to see detected? Leave us a comment and tell us what you think!