MIE: Main Information Enhanced Network for Few-Shot Action Recognition

The goal of the few-shot action recognition task is to recognize actions in videos based on limited action category data. The current main method is to perform multilevel matching on the extracted visual features. However, before feature matching, the richness of the extracted feature information can affect the ability of subsequent feature matching. Therefore, in this article, we propose a feature extraction module with subject information enhancement, which can enhance the network's perception of the main characters and objects in the video during the feature extraction stage, thereby improving the ability to obtain video information. At the same time, we have also redesigned the global-local feature matching method so that the extracted video information can be better matched to complete action recognition. We achieve good performance in three benchmark tests, and also provide a new idea for research in this field.