Actionrecognitionandhumanposeestimationarecloselyrelatedbutbothproblemsaregenerallyhandledasdistincttasksintheliterature.Inthiswork,wepro-poseamultitaskframeworkforjointly2Dand3Dposeestimationfromstillimagesandhumanactionrecogni-tionfromvideosequences.Weshowthatasinglearchi-tectu