Harnessing Object and Scene Semantics for Large-Scale Video Understanding | IEEE Conference Publication | IEEE Xplore