pytorch-cifar
pytorch-cifar copied to clipboard
not a really issue, but a warning
in resnet, self.conv1; the kernel size should be 7 and padding 3 accordingly (if you want to implement the exact same architecture with original resnet). in your implementation, it is 3.
I think considering that the input image size in CIFAR-10 is only 32x32, not using the original kernel size 7x7 in the official resnet could maybe should be more resonable