A novel learning-based frame pooling method for event detection

Lan Wang, Chenqiang Gao, Jiang Liu, Deyu Meng

Abstract: Highlights•A novel learning-based frame pooling method for event detection is proposed.•An effective interpolation operation is proposed to handle the problem of different lengths of different video clip samples.•The pooling weight learning is formulated as an optimization problem and the best weights can be automatically learned.•Extensive experiments show that our method has significant improvements than traditional methods on a challenging dataset.