1 paper · 1 filter
Fengshun Wang, Qiurui Wang, Yuting Wang
The present few-shot temporal action localization model can't handle the situation where videos contain multiple action instances. So the purpose of this paper is to achieve manifo…