Augmented reality app Blippar has been around since 2011, but until recently it focused mostly on advertising and content for brands: Point your Blippar smartphone app at a bold “B” embossed on the pages of a magazine or a bottle of ketchup and more information would pop out on your phone’s display.
But it’s safe to say that augmented reality is coming into a new phase: The contextual information being supplied is getting smarter, and people are gradually becoming more aware of the capabilities of AR and virtual reality (some are even excited to wear headsets, if you can believe it). So Blippar, in an effort to evolve along with the rest of the AR world, has just launched a new version of its smartphone app that is supposed to recognize literally any object you point at it — whether it has been “tagged” with an AR code or not.
Blippar co-founder and CEO Ambarish Mitra showed off the new version of the mobile app today at the Code/Media conference at The Ritz-Carlton, Laguna Niguel in Dana Point, Calif. He pointed the app at a variety of random objects — a magazine, a salad and an apple — to demonstrate how the app’s image recognition capabilities work.
“This is a really big change in our business model,” Mitra had said in an interview before the conference kicked off. “Initially, AR was about very static image recognition. You store images of Starbucks or Coca-Cola or General Mills in our database, and the images match. But now you’re able to analyze any environment in the world in real time, over a 3G connection.”
Mitra said over the past year and a half he has moved his technology team from the U.K. to Mountain View, Calif., to focus on machine learning, which is all the rage in Silicon Valley right now, with everyone from small upstarts to behemoths like Google trying to crack the code on how to make accurate predictions from large sets of data. (Google, actually, has an app that works similarly called Google Goggles, but it works when you point the app at a QR code or a famous landmark or something else recognizable — not necessarily everyday objects.)
In short, this is not an easy thing to do. In fact, ahead of the event, one of our staffers tried it out by pointing the app at his dog, and it thought the pup was a goat.
Mitra has said that, right now, the technology has elementary capabilities, like the brain of a six-year-old; it can recognize “car,” but not “Prius,” or it can recognize an item of clothing, but not the label. However, with machine learning, the app should be able to get to the level of an 18-year-old pretty quickly, Mitra said, in terms of its recognition abilities. And during the onstage demo, it did properly identify a pug named Milton as a dog. (No dogs were harmed in the making of Code Media.)
This article originally appeared on Recode.net.