Google’s Clips camera may be a little creepy, but it also seems pretty useful for the right user; deploying machine learning to automatically snap the best pictures of your kids and pets. But the key to that functionality isn’t just Google’s AI prowess, it also requires a specialized processor built by the Intel-owned chipmaker Movidius.
The chip in question is the Myriad 2, which Movidius describes as a “visual processing unit” or VPU. (That’s as opposed to a graphics processing unit, GPU; or central processing unit, CPU.) The Myriad 2 is a processor tailor-made to handle machine vision tasks like object recognition, and Movidius claims it’s the “industry’s first always-on vision processor.” It’s previously shown up in Google’s Project Tango devices as well as DJI’s autonomous drones, and helps to make their on-board vision processing more efficient.
Google has long been interested in Movidius’ chips. As well as using their VPUs to power Project Tango, the search giant embarked on a partnership with Movidius last year to improve how image recognition works on devices like smartphones. With the launch of the Clips camera, we have a perfect example of the sorts of benefits these collaborations bring.
Clips does it all its AI processing on-device rather than relying on a connection to the cloud to scan images for familiar faces. That’s good for user privacy (there’s no chance of data being snaffled in-transit), but also increases the device’s battery life (because it doesn’t have to maintain an internet connection at all times). These benefits are the direct result of using a specialized chip like Movidius’ VPU.
More and more companies are turning to tech like this to improve on-device AI. Only last month Apple unveiled new iPhones complete with dedicated AI “neural engine” processors, and Huawei showed off similar capability with its recent Kirin 970 chipset. On-device AI is the future and specialized silicon is helping deliver it.