Yes, referring to that.
The layers are defined like:
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = norm_layer(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = norm_layer(planes)
If its just about getting the right shape of out, one could replace it with a single convolution:
conv = conv3x3(inplanes,planes,stride)
no?
But you mean the fact that there is a nonlinearity ReLU in between changes that?
Best, JZ