The output of features is 512*7*7,why we still need AdaptiveAvgPool2d here to make the output size 7*7 output diamension

[shenlinyao]

vision/torchvision/models/vgg.py

Line 44 in 13b35ff

x = self.avgpool(x)

[fmassa]

Because some times you might want to pass images of different sizes, but still want the final classifier to work.
See #643 and linked issues for more context