Original PyTorch example for CIFAR-10 results in 57 % accuracy

I thought it was my own code that is problematic (see this post: CIFAR-10 bad results, loss: 1.1991, Accuracy: 6296/11001 (57%))

I run the original example (http://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html), without any modifications, this is the result:

Accuracy of the network on the 10000 test images: 57 %

Why are the results this bad?

because we use a small network, dont optimize for very long, probably.

this works very well.

1 Like