# of parameters in BatchNorm layer?

In the Week2 assignment, this operation is performed after conv2d;
Batch Norm - 64 filters - 7x7 filter size - (2,2) stride.

Output says it has 256 parameters, but how did we get that number? If we consider beta and gamma as the learnable parameters per filter, shouldn’t we get 64*2=128 parameters?

Hello @DHAiRYA,

This is how we can check this kind of thing:

  1. The doc:

  2. The code:

So, there are not just gamma and beta.


Got it, thank you @rmwkwok!