I\'m training a Resnet18 in pytorch using the pre-trained weights as initialisation. After that I test the performance on a separate test-set. What I noticed is, that the t
