A seemingly endless stream of the boringly mundane mixed with the absurd. This slideshow, never intended for a human audience, exposes the logic by which the data was compiled.
The COCO dataset consists of 328,000 photos depicting objects in 91 categories. The authors downloaded images from Flickr using combinations of COCO classes as search terms. ‘Common objects’ are put ‘in context’ not by any natural circumstance in which they might appear, but by their appearance together with other categories: an often absurd assemblage, uncommon and out of context.