A joint model for action localization and classification in untrimmed video with visual attention | IEEE Conference Publication | IEEE Xplore