Why FrozenBN ?

[fightingaaa]

thanks for this great work!!! but I have a question, in the backbone ,you use FrozenBatchNorm3d:

self.running_mean.zero_() self.running_var.fill_(1) set the running_mean to zero,var to1.

but why not keep running_mean and running_var the same as pretrained models? make this module.eval(). waiting for answers, Thanks!!!