Please download the dataset CIFAR10 and randomly select 1000 images from the training set. We consider that you have the annotation on these images. You can also select 5000 images equally balance from the training set. This set will be the validation set.
Finally, the rest of the official training set are images without labels that you can use to do whatever you want.
Hence you have four sets.
one training set with the labels (1000 images)
one validation set with the labels (5000 images)
one testing set with the labels (10000 images) (the official testing set)
one set compose of all the images that you did not select from the official images from CIFAR10. This set is the unlabel set.