Table 4.

Comparison of YOLO and its successors

Yolo version	Architecture (Backbone)	Neck	Anchor boxes	Features
Yolo(v1)	Darknet (GoogLeNet 24 layers)	–	Handpicked	1 × 1 convolutions, global average pooling, linear activation and leaky ReLU.
Yolo(v2)	Darknet-19	–	5 anchor boxes using K-means clustering.	Batch Normalization, high resolution classifier, convolution with anchor boxes, dimension clusters, direct location prediction, fine-grained features and multi-scale training.
Yolo(v3)	Darknet-53	Feature Pyramid Networks	9 anchor boxes using K-means clustering.	Independent Logistic classifiers, multi-scale training and predictions.
Yolo(v4)	CSP Darknet-53	Path Aggregation Network	9 anchor boxes using K-means clustering.	Spatial Pyramid Pooling (SPP), DropBlock regularization, Mish activation, ReLU6, Class label smoothing and Cross Mini-Batch Normalization.