I followed this tutorial http://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html and finetuned a ResNet model. Then I simply saved this model as “torch.save(model, “…”)” and loaded it in my test code as “torch.load(”…")". I got pretty well training and validation accuracy but when I try to test my model, it makes so obvious mistakes that I don’t think it’s about model. I got json labels in the same order with my training folders. And the mistakes are usually like giving the next label instead of the correct one. I checked order of labels and folders, there is no mistake. And I didn’t change the order of folders after training. What can be the reason?
If anyone goes through the same thing, the problem was using uppercase letters for our folder names.