Overview
The CIFAR-10 and CIFAR-100 datasets are designed to facilitate computer vision research. CIFAR-10 consists of 60,000 32x32 colour images in 10 classes, with 6,000 images per class, split into 50,000 training and 10,000 testing images. CIFAR-100 is similar, but contains 100 classes with 600 images each, grouped into 20 superclasses, also with a split between training and testing sets. These datasets are available in Python, Matlab, and binary formats. The data is structured in batches, with detailed specifications provided for each format, including pixel arrangements and label associations. Baseline results are available using convolutional neural networks, with error rates reported under various conditions, including with and without data augmentation and Bayesian hyperparameter optimization. These datasets provide a standardized benchmark for image classification algorithms.
