13.2 Image Classification

Image classification answers one question: what is in this picture? The model takes a whole image and outputs a single label such as 'cat', 'injera' or 'invoice', usually with a confidence score. Modern classifiers are convolutional neural networks (CNNs) that learn layers of features, from simple edges in early layers to whole objects in deeper layers.

A coffee exporter could train a classifier to sort photos of beans into grades. The model never says where each defect is; it just gives the overall label, like 'Grade 1' or 'Grade 3', for the whole image of a sample.