Closed yihan0512 closed 3 years ago
Hi, I'm wondering why you replaced the original conv1x1 with F.pad for downsampling, please see below: https://github.com/lmbxmu/HRankPlus/blob/bcdb51dd7f039d8fa295eeafa7bc78bc7dab3d75/models/cifar10/resnet.py#L68-L84
nvm, I found in the original resnet paper it's option A for the shortcut....
Hi, I'm wondering why you replaced the original conv1x1 with F.pad for downsampling, please see below: https://github.com/lmbxmu/HRankPlus/blob/bcdb51dd7f039d8fa295eeafa7bc78bc7dab3d75/models/cifar10/resnet.py#L68-L84