The Computer Vision Lab

Abstract

To perform tasks such as object detection or face recognition, computer vision algorithms require massive collections of curated and labelled sets of photographs from which they “learn” regularities. To produce these datasets, a large population of precarious workers annotate billions of images to describe their contents to machines whilst barely having the time to look at them. For dataset creators, classification seems to happen in a glance. Closely following the work of key dataset creators, this chapter enquires into the construction of such a viewing subject able to classify in a few milliseconds, tracing its emergence in the computer vision lab, a space where computer scientists experiment with cognitive psychology in order to make human vision tractable to algorithms. The chapter argues that the computer vision lab produces more than concepts about vision; it produces an epistemic configuration for machine vision: a social ontology, an apparatus of labour management and a specific relation to photography where different bodies of knowledge are mobilised, and the contributions of various actors are selectively acknowledged. By analysing the movement of this epistemic configuration from the lab to the industrial environment, the chapter shows how the experiment is already a factory and the annotation environment an epistemic device.