classifier AvgPool2d stride

[Haoyanlong]

The stride in AvgPool2d at model.classifier should be 13

self.classifier = nn.Sequential(
nn.Dropout(p=0.5),
final_conv,
nn.ReLU(inplace=True),
nn.AvgPool2d(13, stride=13)

[alykhantejani]

Hi @Haoyanlong, what model are you referencing?

[yannadani]

@Haoyanlong Are you referring to the squeezenet?

[Haoyanlong]

@alykhantejani @yannadani ,I reference the squeezenet,but I find the code has some difference with package in pytorch. so I debug the code,the output size of nn.Average's last layer is 100066, so to make the output's size 100011, nn.Average(kernel_size=6).Could give me some guild?Thank you!!!

[alykhantejani]

@Haoyanlong what is the input size to your network. I believe the expected input size is 224 x 224 @fmassa @soumith is this correct?

[fmassa]

With #643, squeezenet now uses nn.AdaptiveAvgPool2d, so this can be closed, as it now supports arbitrary input sizes.