My accuracy of CIFAR10 classes isn't same as tutorial

1

I did CIFAR10 tutorial inPytorch.org .
I was expecting to get the same accuracy for each class the same as the tutorial. My network accuracy on the 10000 test images is 54% while it is different for some classes in comparison with the tutorial. Does have any specific reason?

here is accuracy for each class:

Accuracy of plane : 54 %
Accuracy of   car : 44 %
Accuracy of  bird : 40 %
Accuracy of   cat : 50 %
Accuracy of  deer : 37 %
Accuracy of   dog : 24 %
Accuracy of  frog : 74 %
Accuracy of horse : 61 %
Accuracy of  ship : 74 %
Accuracy of truck : 83 %
2

It looks like the code doesn’t use a seed, so these differences are expected.
CC @chsasank what do you think about adding a seed at the beginning?
If it’s useful, I could create the PR for it.

1 Like