CIFAR-10 and CIFAR-100 are labeled image datasets collected by Alex Krizhevsky, Vinod Nair, and Geoffrey Hinton at the University of Toronto, described in Krizhevsky’s 2009 technical report “Learning Multiple Layers of Features from Tiny Images.” Both are labeled subsets drawn from the larger “80 million tiny images” collection. Each consists of 60,000 color images at 32-by-32 pixels.
CIFAR-10 sorts its 60,000 images into 10 mutually exclusive classes - airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck - with 6,000 images per class and a fixed split of 50,000 training and 10,000 test images. CIFAR-100 uses the same 60,000 images but spreads them across 100 fine-grained classes (600 images each, 500 for training and 100 for testing), grouped into 20 coarser superclasses. The two datasets let researchers test methods at two difficulty levels with identical image format.
Because the images are small but full color and genuinely varied, CIFAR sat between the toy simplicity of MNIST and the scale of ImageNet. It became one of the most common benchmarks for convolutional networks, regularization techniques, and architecture search - the same group around Hinton that built CIFAR would, three years later, win ImageNet 2012 with AlexNet. For business readers, CIFAR is a reminder that the cheap, well-curated public benchmark is often the quiet infrastructure that lets a whole field measure progress and move fast.