1 paper · 1 filter
Tao Wu, Shuqiu Ge, Jie Qin +2
Spatio-temporal action detection (STAD) is an important fine-grained video understanding task. Current methods require box and label supervision for all action classes in advance.…