VGG19 avgpool layer before classifier

vision
1

Hi there
I am new here.
I have a quick question. When I load the pretrained VGG19 model in python and print the model I noticed that there is a avgpool layer before the classifier layer and this doesn’t seem to appear in the original VGG structure sheet:

image
image996×1010 244 KB

Can anyone explain why the avgpool layer is either not mentioned in the VGG structure or why it has been added to the pytorch model?

Best regards
Martin

2

Found this in the original paper
“We then perform global average pooling on the resulting feature map, which
produces a 4096-D image descriptor. The descriptor is then averaged with the descriptor of a horizontally flipped image.”

But I am still confused why such an impactful layer is not displayed in the structure sheet.