Amazon’s $249 deep learning-enabled video camera is now available for developers.
First unveiled last fall, AWS DeepLens is a programmable video camera designed to teach the basics of deep learning through example projects and tutorials. Developers of all skill levels can get started “in less than 10 minutes,” Amazon says.
Deep learning, for the initiated, is a complex machine-learning technique that maps inputs to outputs by finding common patterns in labeled data and using the knowledge to categorize other data samples. Feed a deep-learning application enough pictures of cats, for instance, and it will be able to detect whether a photo contains a cat.
AWS DeepLens “runs deep-learning models directly on the device, out in the field,” AWS Chief Evangelist Jeff Barr explained in a Wednesday blog post.
Sample projects let you train the device to “classify your food as either hot dog or not a hot dog;” identify objects like pants, pens, and headphones; detect cats, dogs, or human faces; transfer the style of one image to a video sequence in real time; recognize actions like playing the guitar; and more.
Amazon recently held a hackathon challenging participants to build machine-learning projects utilizing AWS DeepLens. The winning project, created by a father and senior software engineer named Alex Schultz, who had no prior machine-learning experience, is a deep learning-enabled application called ReadToMe that can read books to kids. You simply show the AWS DeepLens device a page, and it will read it out loud.
As for specs, AWS DeepLens features a 4MP camera capable of capturing 1080p video, 2D microphone array, Intel Atom processor, dual-band Wi-Fi, USB and micro HDMI ports, and 8GB of memory. It runs Ubuntu 16.04 and comes pre-loaded with the AWS Greengrass Core as well as device-optimized versions of Apache MXNet and Intel’s Compute Library for Deep Neural Networks.