AVA: A Finely Labeled Video Dataset for Human Action Understandingresearch.googleblog.com44 pointshurrycane9 years ago