ImageNet, launched in 2009, became a major benchmark for image-recognition systems.
ImageNet is a large visual database organized according to concepts from WordNet, a lexical database of English. Its images are grouped into labeled categories such as animals, objects, and places, giving researchers a common way to train and compare computer-vision systems.
The ImageNet Large Scale Visual Recognition Challenge made the dataset especially influential. In the competition, systems classified images and located objects across thousands of categories. The sharp improvement in results during the early 2010s helped draw attention to deep convolutional neural networks and large-scale supervised learning.
ImageNet is sometimes confused with MNIST, which contains small handwritten-digit images, or COCO, which emphasizes everyday scenes and object captions. ImageNet’s role was broader large-scale visual classification, although researchers have also discussed dataset bias and the limits of using benchmark accuracy as a complete measure of intelligence.