Name what is in the picture

The competition's central task was simple to state: give a system a picture and ask it to name the main object.

A picture of a school bus should produce the label “school bus.” A picture of a coffee mug should produce “coffee mug.” Turning an image into one category label is called image classification.

Here, correctly recognising an image had a narrow meaning. The system did not need to describe everything in the scene. It had to choose the right category from a set of 1,000 possibilities.

The challenge supplied about 1.2 million labelled pictures for training. Competing systems then faced the same separate test pictures, with the correct labels hidden.

That shared task, data, and scoring rule formed a . Like an exam, everyone received the same questions without the answer sheet.