The CIFAR-10 dataset is a labeled subset of the 80 million tiny images dataset consisting of 10 classes. They were collected by Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton:

• Source (CLASSES)

• Original images (162MB, contains Python pickled objects)

• Image classification dataset: train (120MB), train (24MB)

Conversion from original data:

• Decompress the images

• Run Python script (requirements.txt) from the directory with the .tar.gz file

• The train and test directories contain training and test sets with the the class labels as sub-directories

License

Unclear.

Citation

Learning Multiple Layers of Features from Tiny Images, Alex Krizhevsky, 2009.