You watch once more: a more effective CNN architecture for video spatio-temporal action localization.

Efficient abnormal behavior detection with adaptive weight distribution.