Feature Fusion Methods in Deep-Learning Generic Object Detection: A Survey
Jiang Deng, Bei Sun, Shaojing Su, Zhen Zuo
Abstract
Feature fusion has become one of the most popular orientations in object detection, which has been widely applied to enrich object representation, especially for the small objects. However, there has not been published an integrated survey paper that concentrate on the feature fusion methods in deep-learning object detection. Therefore, we would like to sort out the relevant content. And we believe that a comprehensive survey is necessary and can provide some useful guidance for the follow-up work. In this paper, we first introduce some classical backbone networks which adopt feature fusion methods. Then we analyses the fusion techniques of several typical or state-of-the-art frameworks. Thirdly, we present a synthesize survey of fusion strategies. At last, the future development trends and challenges are summarized. This survey infers that feature fusion methods have acquired some good results especially in recent five years, but further improvements or potential research directions still widely exist.