Temporal Feature Enhancement Module for Action Recognition

Abstract

Although convolutional neural network has achieved great success in the field of image classification, there are still problems to be solved for video action recognition. Compared with image classification, temporal modeling remains key and challenging for action recognition in videos. To mitigate this issue, this paper presents an action recognition network module, termed as Temporal Feature Enhancement module, with a focus on enhancing the use of temporal feature to achieve efficient action recognition. The module is composed of a Short-term temporal feature enhancement module and a Long-term temporal feature enhancement module. These two modules work together to alleviate the problem of temporal feature modeling in action recognition networks. Through experiments on UCF-101 and HMDB-51 datasets, the proposed algorithm is superior to the comparison algorithms, which proves its effectiveness.