Better deep visual attention with reinforcement learning in action recognition | IEEE Conference Publication | IEEE Xplore