Paper Title (use style: paper title)

Main Part

1. YOLO

You Only Look Once (YOLO) is an open-source object detection algorithm based on convolutional neural networks and it was first introduced by Redmon et al. In 2016. YOLO is one of the most well-known deep learning algorithms, and it stands out with its speed due to its single-stage detection architecture “Fig. 1” []. Detection systems prior to YOLO reuse classifiers or localizers for object detection.

Fig. 1. Architecture of YOLO algorithm.

Firstly, YOLO algorithm divides the input picture into grids of MxM for detection. The sizes of these grids may differ from one version of YOLO to another, for example, grids such as 3x3 5x5 19x19 can be used. With the help of each grid within itself, whether there is an object in the field or not can be find out, if it is in its midpoint, if it is within its midpoint, its length, height, and class. After performing these operations, bounding boxes are created. Then, an estimation vector is will be created for each grid. Within the prediction vector are the confidence score, Bx (x coordinate of the object's midpoint), By (y coordinate of the midpoint of the object), Bw (the width of the object), Bh (the height of the object) and the dependent class probability [2].

2. YOLOv5

YOLOv5 is a new version of the YOLO family of object detection algorithms. It was released a month after YOLOv4 and is different from previous releases in several ways. Firstly, YOLOv5 uses PyTorch instead of Darknet. This makes it easier to use and customize, and it also allows for faster training and inference. Secondly, YOLOv5 uses CSPDarknet53 as its backbone network “Fig.2”. This network is more efficient than the Darknet53 network used in YOLOv4, and it also helps to improve accuracy. Thirdly, YOLOv5 uses Path Aggregation Network (PANet) as its neck. PANet helps to improve the flow of information through the network, which can lead to better object detection results. Finally, YOLOv5 uses the same head as YOLOv4 and YOLOv3. This head generates three different output feature maps to achieve multi-scale prediction. Overall, YOLOv5 is a significant improvement over YOLOv4. It is more efficient, more accurate, and easier to use.

Fig. 2. Architecture of YOLOv5 algorithm

3.