ImageNet is a database of annotated images created by researcher Fei-Fei Li with Stanford University and Princeton University as a resource for research into artificial vision. Its corpus of words comes from WordNet, a lexical database for the English language organised hierarchically on the basis of the meaning of words. ImageNet has compiled a large number of images for each of the categories, thereby transforming WordNet into a kind of visual dictionary.
However, words may have a meaning but this is not to say that they have a visual equivalent. The limits of the connection between images and words are plain to see if you look through the ImageNet categories relating to people: is it possible to create a dataset of images for the category of a ‘bad person’? Are moral criteria elements that are visible in images? In the case of gender, the ImageNet images are also telling: for example, the ‘smasher, stunner, knockout […]’ category (“a very attractive or seductive looking woman”, according to the definition of the dataset itself) does not define the content of the image but the type of gaze with which it has been constructed. The rhetoric of images -police rhetoric in one case, sexual in the other- is ignored in both examples.