Google's 'Move Mirror' AI Matches Pose With A Catalog Of 80,000 Images

AI can do a lot of things. And Google that is known as one of the heavyweight supporter, is having a bit of fun.

The company has published an AI experiment called Move Mirror. Just like the name suggests, it matches users' pose. And not just that, as the AI "mirrors' the users with a catalog of 80,000 photos, all in real-time. What this means, when the body moves, so does the match returned by Google.

To use the AI, users just need to visit the Move Mirror website and grant it access to their computer’s webcam.

Stand where the webcam can see, and the AI will map the positioning of users' joints using PoseNet, an computer vision model that detects human figures in images and videos by identifying where key body joints are.

The AI then contrasts that pose against Google's massive library, and returns with the one that (closely) matches the pose. And not just that, users can also create a GIF in the process.

To make this happen, Google's Move Mirror uses a JavaScript library called 'TensorFlow.js', an open-source library where anyone can use to define, train, and run machine learning models entirely in the browser, What this means, all the heavy computing happens inside users' computer, and no data about users' pose is sent to Google.

"Your images are not being stored or sent to a server," assured Google on its blog post.

"With Move Mirror, we’re showing how computer vision techniques like pose estimation can be available to anyone with a computer and a webcam. We also wanted to make machine learning more accessible to coders and makers by bringing pose estimation into the browser—hopefully inspiring them to experiment with this technology."

The progress of AI and machine learning has been fascinating. With better hardware and software, they can be created to be better in many things, like detecting objects in images, helping to detect diseases, and even enabling cars to drive themselves.

But AI can also be used in more playful ways, as Google demonstrates

Published: 
20/07/2018