InstanceNorm3d in training and testing!

In the in InstanceNorm3d document

  >>> # Without Learnable Parameters
 >>> m = nn.InstanceNorm3d(100)
>>> # With Learnable Parameters
>>> m = nn.InstanceNorm3d(100, affine=True)

So, which one is for training and which setting is for testing

The training/evaluation mode is switched via model.train() and model.eval(); The affine argument does not have anything to do with training vs. evaluation mode determination. Affine just flags wether to apply affine transformation or not.

1 Like

Thanks. So it is similar batchnorm setting. However, the performance is worst than batchnorm