Mean / Variance Dimensions of Window Norm

[L-Reichardt]

Hello,

I have a question about window norm.

Are the mean / variance calculated over the channel dimension as well, so in a tensor (N, C, K) -> Mean = sum(C, K)/(C*K)? Or is this calculation just performed for each channel individually per neighbor -> Mean = sum(K)/K ?

Best Regards,
L

[DBDXSS]

(N, C, K) -> Mean = sum(C, K)/(C*K) is right.