Is IBN = MVN_layer in caffe？

[math-yyj]

Is IBN layer just "MVN_layer in caffe"？

Thank you very much!

[XingangPan]

Instance normalization (IN) layer is "MVN_layer in caffe" plus a scale and shift operation.
IBN layer is half IN half BN.

[math-yyj]

I see, so it should be MVN_layer + Scale_layer

many thanks again ~

[vd001]

I tried to replace the Instance normalization (IN) layer with "MVN_layer + Scale_layer" in caffe, but found the network hard to converge. When i remove every scale layer following MVN(i.e. use MVN layer only), the network converges.
My question is:

1. If i replace IN with MVN layers only in caffe, does it hurt the generalization or transfer ability of IBN-Net, or is the scale layer really important?
2. what makes the net hard to converge when IN layer is followed by scale layers?

thanks!