It is fine to eval net with BN layers with batchsize > 1?
Yes. During evaluation (after model.eval()
was called), the running estimates for the mean and variance will be used for every sample, so the batch size doesn’t matter.
It is fine to eval net with BN layers with batchsize > 1?
Yes. During evaluation (after model.eval()
was called), the running estimates for the mean and variance will be used for every sample, so the batch size doesn’t matter.