Movidius introduces a USB stick that can quickly implement machine vision into electronics already containing neural network frameworks for machine learning. The USB operates offline, and is expected to accelerate machine vision, changing regular devices into ones that can analyze their surroundings offline.
The stick contains a Myriad 2 machine vision processor unit (VPU)—the same one that is used in DJI’s autonomous drones. Introduced last April as a prototype called Fathom, it was swiftly bought by Intel that following September. Still, the product is produced under the Movidius brand.
Now, the device is being reintroduced into the public at a cost of $79—down from the original $99. Its Myriad processor is low power, operating at a single watt. Twelve parallel cores run vision algorithms like object detection and facial recognition. More than one can be connected for more computing power.
Movidius claims its vision USB can offer up to 10 gigaflops performance, and runs natively on popular neural networks in the Caffe framework. It is not clear if it runs on TensorFlow, which is another popular neural network framework by Google. See more specs here, or watch the promotional video for more information about its performance and a general demonstration of its interface with a microprocessor.