Image classifier

An image classifier is a trained model that takes an image as input and assigns it to one of several categories set in advance, such as “cat” or “dog.” The training algorithm sets the parameters using thousands of already-categorized images; no human decides in advance which image features tell the categories apart.

To the model, a photo is a set of numbers: the brightness and color values of its pixels. Its immediate output isn’t a word but a score for each category. A separate step then picks the highest.

An example: A classifier knows the categories cat, dog, fox and car. For a photo of a cat it outputs something like cat 6.2, dog 2.9, fox 1.4 and car −3.0 (made-up values), and the screen shows “cat.” It knows no other answers, so even a photo of a horse lands in one of the four: the one with the highest score.

Not to be confused with an image generator: An image classifier sorts existing images and creates no new ones. It’s closer to a trained spam filter: both sort an input into a fixed list of categories, just with images instead of emails as training data. A language model also picks from a fixed list, its vocabulary, in every round. New text only comes about because the chosen piece is appended and the loop repeats.

Where you’ll come across it: You’ll more often read “image recognition” than the technical term. The idea is at work in apps that name a plant or an animal from a photo.

More on this in “Input and Output: What a Function Does”.