Table 2.

The table depicts the number of convolution, pooling layers, FC layers, and basic features of AlexNet, VggNet-16, ResNet-18, and VggNet-19 architectures

Model	Year	No. of Covoluti on Layers	No. of Pooling Layers	Fully Connect ed Layers	Main Features
AlexNet	2012	5	3	3	▪ First CNN architecture to win ImageNet challenge with top-5 error rate of 15.3%.
					▪ Used ReLU as activation function instead of tanh or sigmoid.
					▪ AlexNet has 60 million parameters.
					▪ It had used the Stochastic Gradient descent as the learning algorithm.
VggNet-16	2014	13	5	3	▪ The model achieved 92.7% top-5 test accuracy in ImageNet challenge.
					▪ The model replaces the the large sized kernals used in AlexNet with 3✕3 sized multiple kernals enabling better learning.
					▪ Main con of this network is that it is slow to train.
					▪ And network architecture weights are quite large.
ResNet	2015	17 with 8 residual units	2	1	▪ Main building blocks are residual blocks that increase the performance of the network.
					▪ The identity connection helps the network to handle vanishing gradient problem.
					▪ The batch normalisation used by network mitigates the problem of covariant shift.
					▪ ResNet 18 has residual blocks of two layers deep.
VggNet-19	2014	16	5	3	▪ Has 3 additional convolutional layers than that of Vgg-16.
					▪ Deep network is believed to train more efficiently.
					▪ Requires more memory than Vgg-16.