An Improved Object Detection of Image Based on Multi-task Learning

Object detection has a wide range of applications in the real world. For example, accurate identification of object in autonomous driving is an important prerequisite for realizing safe autonomous driving. The current object detection models can't utilize more detailed semantic information and solve the problems of different objects in the image. This paper improves YOLOV5 by introducing semantic segmentation to structure a multi task framework. Convolutional Neural Network (CNN) is used to extract features from input image in this paper. The feature maps are shared by detection thread and segmentation thread to complete object detection and semantic segmentation. Experimental results on the Cityscapes dataset show that the improved model achieves better performance compared with some other models.