How to use the ResNet_SSN as a pretrained network?

[Ning5195]

Nice job!
I want to use the ResNet_SSN as a pretrained network for the detection. However, as the 'mean_fixed' and 'var_fixed' are '1', the parameters of 'mean_weight' and 'var_weight' will not change during the training. What should I do if I want to get the specific parameters for the detection task?
What's more. I try to set the 'mean_fixed' and 'var_fixed' to '0', and the initial 'mean_weight' and 'var_weight' to '1', but I get a worse results than using the parameters of the pretrained model. For example, the data of 'mean_weight' is [9.9426e-01, -3.2573e-06, -2.9768e-06]. But the data of 'mean_weight' in the pretrained model is [1, 0, 0].
I'd appreciate your help.