C4W2E1 identity_block

I don’t really understand why filter shape of F1 and F3 is (1,1) and for F2 is (f,f).
Is there any particular reason applying it for the implementation of ResNet block?

It allows you to vary the f value without rewriting the function.

I mean why we choose these kernel dimensions (e.g. bottleneck) ?

Experiments by the course designers showed that these values worked well for what the assignment is teaching.

1 Like