Next, we see the architecture of the VGGNet, specifically the VGG-16 flavor that contains, unsurprisingly, 16 layers. All convolution layers have filters of spatial size 3x3, and the number of filters in the convolution layers increases from 64 to 512 as we go deeper into the network.

The simple modular design of stacking two or three convolutional layers followed by pooling allows the network to be easily increased or reduced in size. As a result, the VGG successfully created and tested versions with 11, 13, and 19 layers: