Representing Videos Using Mid-level Discriminative Patches | IEEE Conference Publication | IEEE Xplore