Breaking video into pieces for action recognition