Lightweight One-Stage Object Detection Based on Deep Learning

Abstract

Object detection is a basic and challenging problem in the field of computer vision. In recent years, with the development of deep learning, models using deep networks can automatically extract image features, and they are also good at feature representation. These models have dramatically improved accuracy and efficiency. Thus, object detection based on deep learning has become the main processing method in the field of vision. This thesis proposes a lightweight one-stage object detector based on deep learning. The main contributions of the proposed method are two-fold: (1) We design a rescaled pyramid network module to generate the different level feature maps. (2) We lightweight an existing detection head network. The performance of the proposed method has been evaluated based on the experiments on MS COCO dataset. Experimental results show that the proposed method achieves 20.1 mAP, surpasses prior lightweight one-stage detectors.