Scaling in all layers

[Jayan-K-Duggal]

I want to know how you are implementing scaling in all the layers in Pytorch, originally which is being done in Caffe model in each layer.

[luuuyi]

@JK009 if the scaling you mentioned is like y = k*x, it can easily implement in pytorch, only write y = k * x can finish this work, x and y is tensor meanwhile k is a scala.

[Jayan-K-Duggal]

I want to rephrase my question and ask have you done scaling in the pytorch code of Squeezenext.

[Jayan-K-Duggal]

is transform.resize equivalent to scale layer, is meant for smooth transition from input to our convolution, did you implement this in your squeezenext code or not ? output = F.relu(self.bn1(self.conv1(input)))
I don't see any scaling layer in your code.This makes it a little different from caffe implemented base squeezenext arch or not? Please reply asap.