Distilling interaction knowledge for semi-supervised egocentric action recognition

Haoran Wang, Jiahao Yang, Baosheng Yu, Yibing Zhan, Dapeng Tao, Haibin Ling

Abstract: Highlights•We distill the hand–object interactions with limited annotated videos.•We adopt the hand–object interactions to predict pseudo-labels of unlabeled videos.•The experimental results show the effectiveness of the proposed method.