DeepMind, a London-based artificial intelligence research lab, has published a database called Kinetics consisting of over 300,000 high-quality YouTube clips depicting human actions to help train AI systems to identify human tasks. The clips exemplify one of 400 different classes of human actions, such as playing an instrument or shaking hands, and each class of action has 400 example clips. DeepMind created Kinetics as a dataset for the ActivityNet Large Scale Activity Recognition Challenge, a competition to encourage AI developers to improve how AI systems understand human activity.
Image: PublicDomainPictures.