TS-YOLO:An efficient YOLO Network for Multi-scale Object Detection
Yang Wang, Bo Ding, Li Su Tong
Abstract
To solve the problem that the You Only Look Once(YOLO) v4 still has missing detection in multi-scale object detection, we proposed a novel deep convolutional network structure TS-YOLO with three spatial pyramid pooling(SPP) modules in YOLOv4. SPP plays an important role in multi-scale object detection, it can extract more semantic information in complex scenes. In this paper, we add two more SPP modules and redesign the pooling core sizes in SPP on the basis of YOLOv4. Our training was on the Pascal VOC data set and the experimental results show that our TS-YOLO not only detects more objects but also has 2.21% higher accuracy compared with original YOLOv4, demonstrating the excellent performance of our model in multi-scale object detection.