Mastering Computer Vision with TensorFlow 2.x

YOLO versus YOLO v2 versus YOLO v3

A comparison of the three YOLO versions is shown in this table:

	YOLO	YOLO v2	YOLO v3
Input size	224 x 224	448 x 448
Framework	Darknet trained on ImageNet—1,000.	Darknet-19 19 convolution layers and 5 max pool layers.	Darknet-53 53 convolutional layers. For detection, 53 more layers are added, giving a total of 106 layers.
Small size detection	It cannot find small images.	Better than YOLO at detecting small images.	Better than YOLO v2 at small image detection.
		Uses anchor boxes.	Uses a residual block.

The following diagram compares the architectures of YOLO v2 and YOLO v3:

The basic convolution layers are similar, but YOLO v3 carries ...