Systematic Advancement of Yolo Object Detector For Real-Time Detection of Objects
Ejiyi Chukwuebuka Joseph, Olusola Bamisile, Nneji Ugochi, Qin Zhen, Ndalahwa Ilakoze, Chikwendu A. Ijeoma
Abstract
This paper explicates the systematic advancements that were observed from the inception of the YOLO (You Only Look Once) object detector to the most recent version 4. Since its introduction in late 2015, YOLO has recorded tremendous implementation as well as improvements and applications. In this work, a brief survey of the YOLO network is presented considering the introduction that was made to each version that succeeded each preceding version and the advancement on how the model performed with detection. We used the latest version of the network (YOLOv4) to train 50 classes of objects that we considered popular objects for real-time detection. The model trained obtained an mAP of 64.80% @IoU of 0.5 and when deployed for real-time detection, it achieved a 43FPS speed of detection.