Much of the pioneering work on deep learning in computer vision focuses on image classification using the ImageNet dataset.
Most methods reshape the input images to a standard size. In a typical system, the input to the network is a RGB image, and the output is a probability distribution over the 1000 classes. ImageNet is quite challenging because there are a large number of classes, exhibiting considerable variation. Before deep networks were applied, the state-of-the-art method classified the test messages with errors for the correct class being in the top five suggestions. Five years later, deep learning models eclipsed human performance.
AlexNet was the first method to do well on this task.