texture_nets icon indicating copy to clipboard operation
texture_nets copied to clipboard

The shape of running_mean

Open wei61547-jp opened this issue 3 years ago • 0 comments

Why the shape of running_mean in Instancenorm is [C] not [N,C] ? the code " self.running_var:copy(self.bn.running_var:view(input:size(1),self.nOutput):mean(1))" I don't the reason running_mean/var view to (N,C) and mean(1)? So it similar to Batchnorm? I find the cuda-kernel of Instancenorm used batchnorm's kernel ,the shape of running_men in Bn is [C], Instance's also [ C]too?

wei61547-jp avatar Oct 14 '21 08:10 wei61547-jp