Video2vec: Learning semantic spatio-temporal embeddings for video representation | IEEE Conference Publication | IEEE Xplore