For machines, it’s not exactly easy to seamlessly copy bodily positions of humans or even a static pose. However, thanks to machine learning and Move Mirror AI, it can now be done. Using it, tech giant Google has come up with a process that would enable computers mirror every human move perfectly.
Not only can it copy perfectly a static pose, but also moving parts. The team of scientists at TensorFlow of Google has managed to achieve the feat in an experiment on AI known as Move Mirror AI. It makes use of Google’s PoseNet neural network to uncover poses and then attempts to copy those.
PoseNet, in turn, banks upon machine learning to recognize the human body’s position by unraveling the positions of various parts and joints in a video or photo.
Model Matches Pose with 80, 000 Images
The model then goes on to match it with a set of almost 80, 000 photos, revealed a creative technologist at Google Creative Lab in a blog post. According to the post, the machine learning model is computed either in the browser or on-device. Google, however, does not store the images or transfer them to a server.
One needs to move before the front camera of a phone or webcam so that Move Mirror can copy every movement. It is akin to a mirror that can match your moves with those of people from the domain of sports, acting, and others. The end result can be in the form of a GIF which can be shared with friends.