VLAD3: Encoding Dynamics of Deep Features for Action Recognition | IEEE Conference Publication | IEEE Xplore