Deep snippet selective network for weakly supervised temporal action localization

作者：

Highlights：

• We propose a DSSN model, which is a novel weakly supervised method for temporal action localization.

• We utilize a ternary mask to erase the most discriminative zones and mark the background zones simultaneously.

• We present a background suppression branch to further improve the accuracy.

摘要

•We propose a DSSN model, which is a novel weakly supervised method for temporal action localization.•We utilize a ternary mask to erase the most discriminative zones and mark the background zones simultaneously.•We present a background suppression branch to further improve the accuracy.

论文关键词：Weak supervision,Temporal action localization,Erasing branches,Ternary mask,Background suppression branch

论文评审过程：Received 8 November 2019, Revised 6 September 2020, Accepted 2 October 2020, Available online 2 October 2020, Version of Record 1 November 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107686