batchnorm about the mxnet model

[hnuhchen]

Hello

I have a question about your code. In the fiel json2prototxt.py, you transform batchnorm layer of mxnet to batchnorm layer and scale layer of caffe, and exactly the bias of scale layer is true. But in this situtation, the batchnorm layer has 4 learning parameters but the scale layer has no parameter. So what does the scale layer do?

Or could you provide you mxnet model with batchnorm layer which pass the transform test?

Thank you