Announcing AVA: A Finely Labeled Video Dataset for Human Action Understanding

research.googleblog.com