Google Clips camera comes with an Intel Movidius Myriad 2 VPU which powers the on-device AI processing

Google announced an interesting product at its 'Made By Google' hardware event on 4 October alongside the Pixel 2 / Pixel 2 XL, Google Home Mini and Max, the Pixelbook, Pixel Buds and more. This hardware product called Google Clips captures seven-second clips without any human intervention and employs on-device artificial intelligence processing to do that.

Image: Google

Image: Google

Google’s camera has a traditional shutter button. But users may also clip the camera onto an object and have Clips automatically record “stable, clear shots” of familiar faces set by the user, the company said. Each shot lasts seven seconds, and the battery survives about three hours in the smart-capture mode, according to Google. Not only does Clips capture images and videos, but it also lets selects and presents the best ones to you based on the on-device AI processing.

This is made possible thanks to the Intel Movidius Myriad 2 vision processing unit, which does the on-device artificial intelligence processing. According to Intel, the on-device processing reduces the overheads that would be required for cloud processing of the same data. This makes the Clips' hardware more power efficient and reduces latency. Also, you will not require internet access to use the camera.

"In our collaboration with the Clips team, it has been remarkable to see how much intelligence Google has been able to put right into a small device like Clips," said Remi El-Ouazzane, vice president and general manager of Movidius, Intel New Technology Group.

According to Intel, the Movidius VPU technology has been developed to enable low-power and high performance computer vision intelligence which can be implemented in intelligent devices.

Addressing some of the privacy concerns around Google Clips, product lead Juston Payne said that Google takes privacy very seriously. "We care very deeply about privacy and control and it was one of the hardest parts of the whole project," said Payne to TechCrunch. According to Payne, till recently what could only be done on the cloud with running neural networks against the data and doing semantic analysis to give you the output, can now be done on-device.

According to TechCrunch, Google went about training the Clips cameras by taking the help of video editors, image raters and asking them to label the clips shot by Google, which would become the training material for Clips. With time the Clips camera will learn to recognise people, and even the animals, you are closest to and what kind of images will interest you. The focus right now for Clips is limited within a home environment, and is not ideal for something like an outdoor vacation. Of course, you can still carry it along on a vacation and take photos using the manual shutter button, but the intelligence part will be missing as the data would be brand new.


Published Date: Oct 05, 2017 12:43 pm | Updated Date: Oct 05, 2017 12:44 pm