UvA-DARE ( Digital Academic Repository ) Objects 2 action : Classifying and localizing actions without any video