A New Architecture of Feature Representation for Object Detection

Abstract

In recent years, excellent object detectors have been working on how to better represent features of multiple scales. In this paper, we propose a new architecture of feature map generation that combines feature pyramid network and context enhancement module. The main contributions of the proposed method are two-fold: (1) We design a pyramid context enhancement module to generate the feature maps for object detection. (2) We address the scale variance problem while rescaling different level features and utilizing global context. The proposed method experiments on PASCAL VOC dataset. Experimental results show that the proposed method can improve the accuracy of baseline by 1.5.