Someone at Google really likes sneakers: The company has just bought a French machine-learning startup that taught a computer how to recognize 15,000 different types of them.
Paris-based Moodstocks builds image and object recognition software using deep learning techniques, and offered an Android app and visual search API that could recognize certain kinds of objects. By analyzing video from a smartphone camera, and correlating it with accelerometer readings to determine how the camera is moving around, the software is able to infer information about the three-dimensional shape of objects in the video, facilitating their recognition.
In February 2015 the company demonstrated its ability to identify sneakers through its app. Three months later, after training the software using 15,000 photos of shoes from an online retailer's website, Moodstocks claimed to be able to shop online for all the sneakers on sale in a Macy's store.
Google has been introducing elements of machine learning into its existing online services, including Google Translate and Inbox, a next-generation interface for Gmail.
Its online photo archival service, Google Photos, uses machine learning to identify categories of photo, such as parties or beach scenes, to make it easier to search.
But there's still a lot of work to be done in this field, according to Google's blog post (in French) announcing the acquisition of Moodstocks.
Google said the Moodstocks team will join its existing research and development operation in Paris.
There, they will develop image-recognition tools for use in Google services, the Moodstocks team wrote on their own site.
Meanwhile, Moodstocks will discontinue its own image recognition services, although paying subscribers will have access until their subscriptions run out, the post said.
Google didn't put a price on the Moodstocks acquisition, but it's unlikely to be as high as the $500 million it reportedly paid in 2014 for the much larger DeepMind, the London-based developer of the Go program that beat top player Lee Se-dol in March.