[enhancement] Replacement nn.AvgPool2d by nn.AdaptiveAvgPool

[Nurfen]

Hello,

Thanks for creating the Resnet in pytorch.
The network could be made more flexible towards different inputsizes when the avgpool2d layer is replaced by the AdaptiveAvgPool layer.

self.avgpool = nn.AvgPool2d(7, stride=1)
could be replaced by

self.avgpool = nn.AdaptiveAvgPool2d(1)

[fmassa]

Hi,

This has been implemented in #643

Thanks for opening the issue!