Abstract
Though action recognition based on complete videos has achieved great success recently, action prediction remains a challenging task as the information provided by partial videos is not discriminative enough for classifying actions. In this paper, we propose a Deep Residual Feature Learning (DeepRFL) framework to explore more discriminative information from partial videos, achieving similar representations as those of complete videos. The proposed method is based on residual learning, which captures the salient differences between partial videos and their corresponding full videos. The partial videos can attain the missing information by learning from features of complete videos and thus improve the discriminative power. Moreover, our model can be trained efficiently in an end-to-end fashion. Extensive evaluations on the challenging UCF101 and HMDB51 datasets demonstrate that the proposed method outperforms state-of-the-art results.