Penn Action Action Recognition Dataset

Penn Action is a human action dataset consisting of consumer videos, and provides action annotations, 2D locations of key points, and visibility for each video frame. The dataset contains 2,326 video sequences involving 15 different actions.