论文部分内容阅读
To solve the task of detecting and recounting events in videos with limited training examples, we propose a novel two-stage hybrid concept temporal pooling approach that is aware of potential concept drift in the video stream. We initially partition vi