How do animated characters dance? New program for machine learning to help
A new machine learning program was recently released in the US that can learn new skills from YouTube videos, such as dance, acrobatic stunts and martial arts moves, which could provide more physically coordinated characters for movies and video games, or serve as a virtual training ground for robots.
Animation and game characters often learn full-body human movement through motion capture equipment, and with this program motion capture becomes easier and cheaper, virtual versions of outdoor activities and the creation of realistic avatars of large animals become more convenient.
The new program utilizes a computer code called an artificial neural network, which roughly mimics the way the human brain processes information. By recognizing about 100,000 photos of people in different poses, the project first estimated the pose of an actor in each frame of a video clip.
These animated characters are still struggling with nimble dance moves, like Gangnam Style jig, and learning from short films with only one person in them.