I gave Cifar10, Cifar100, and SVHN torchvision.datasets a go, and after loading the train_set and test_set, the labels’ identifiers are different:
To get the labels in Cifar10 and Cifar100
test_set.test_labels
train_set.train_labels
In SVHN, the situation is different (and I think more accurate)
test_set.labels
train_set.labels