Spatial misrecognition: The world according to computer vision
Topics:
Keywords: computer vision, AI, deep learning, urban data, inequality
Abstract Type: Paper Abstract
Authors:
Jonathan Cinnamon, University of British Columbia Okanagan
,
,
,
,
,
,
,
,
,
Abstract
At a moment in history when almost nothing escapes the watchful eye of the imaging sensor, an ongoing but accelerating shift in the role of the visual image is underway. Simply, an image is no longer a mere representation of the world; instead – leveraging advancements in computer vision – every digital image is now a dataset enabling action in the world. Critical attention has been paid to computer vision in the context of generative learning and facial recognition systems, however less attention has focused on the implications of applying these ‘recognition’ techniques for extracting information from images of urban objects and landscapes. The proliferation of social media, video and photo sharing sites, and street-level image platforms provides a large source of urban imagery to train computer vision models, seemingly circumventing the ethical and political implications of doing so via human biometric image datasets. Yet, through a critical praxis engaging with deep learning computer vision technologies informed by media, cultural, and political studies of urban vision, this paper reveals how these technologies advance distorted spatial imaginaries and provide for the profiling of people according to their geographic identities. This two-fold notion of spatial misrecognition provides a conceptual basis for linking computer vision to the advancement of sociospatial inequalities outside of the image.
Spatial misrecognition: The world according to computer vision
Category
Paper Abstract