Hi All,
I found out the output size of the MaxPool2d could be not consistent with the formula in the document.
My MaxPool2d and the input are declared as:
nn.MaxPool2d(kernel_size=4, stride=4, padding=1)
inputs = torch.randn(1, 1, 1, 1)
According to the output size formula in the document, the output height/width should be:
floor((1 + 2 * 1 - 1 * (4 - 1) - 1) / 4 + 1) = floor(0.75) = 0
However the actual output size is (1x1x1x1)
The PyTorch version is 1.2.0
Does anyone have the same question here? Many thanks!