Support **kwargs for some pytorch args of networks #2164
Description
For networks like SegResNet, we cannot change parameters for normalization layers (such as set track_running_stats=True when we use BatchNorm3d). Therefore, the implementation should be more generic.