A question about fune-tuning VGG16 model

I want to fune-tuning VGG16 model with medical image, and I preprocessd the image by normalization:
def norm(x):
x=x.clamp(max=1500)
x=x/x.max()
for t, m, s in zip(x,[each.mean() for each in x],[each.std() for each in x]):
t.sub_(m).div_(s)
return x
But when I load the parameters and run the net, I found the feature maps become all zeros after the second conv layer, Anyone can help me find the error in my operation?
Thank you !