O que é CIFAR?
CIFAR, que significa o Instituto Canadense de Avançadas Pesquisa, is best known for its collection of datasets used in the field of aprendizado de máquina, particularly in visão computacional. The most popular among these datasets are CIFAR-10 and CIFAR-100.
O CIFAR-10 dataset contains 60,000 32×32 color images in 10 different classes, with each class having 6,000 images. The classes include:
- Avião
- Automóvel
- Pássaro
- Gato
- Veado
- Cão
- Rã
- Cavalo
- Navio
- Caminhão
CIFAR-100, on the other hand, consists of 60,000 32×32 color images as well, but it is divided into 100 classes, each containing 600 images. The 100 classes are grouped into 20 superclasses, which provide a more nuanced categorization of the images.
Both datasets are commonly used for benchmarking the performance of machine learning algorithms, especially redes neurais convolucionais (CNNs), and they serve as a standard testbed for researchers to compare their models. The relatively small size and simplicity of CIFAR datasets make them ideal for rapid experimentation and prototyping in academic and industrial research.
Since their release, CIFAR datasets have become a staple in the AI community, enabling advancements in classificação de imagens tarefas e o desenvolvimento de arquiteturas de redes neurais mais sofisticadas.