Unconstrained College Students dataset
The UnConstrained College Students Dataset (UCSD) is a database comprising 16,000 photographs of approximately 1,700 students going about their lives at the University of Colorado, Colorado Springs, for the research and development of 'face detection and recognition research towards surveillance applications'.
The photographs were taken secretly on 20 different days between February 2012 and September 2013 using a 'long-range high-resolution surveillance camera without their knowledge,' according to Professor Terry Boult, the University of Colorado computer scientist who led the project.
The project was initially funded by the US government Office of Naval Research’s Multidisciplinary University Research Initiatives Program, and later by other US government entities.
The UCSD was first reported by the Colorado Springs Independent in May 2019, a month after the dataset had been taken down. The expose ignited a firestorm amongst the local media, which focused on its intrusiveness and opacity. It also faced criticism from a University of Denver law professor.
Shortly afterwards, the Financial Times cited the project as an example of an attempt to gather personal images to improve facial recognition systems in as natural (ie. 'wild') a manner as possible - ideally covertly.
'Even [LFW] photos aren’t that wild because people know they are being photographed and uploaded on the internet. But these are students walking on a sidewalk on campus, who are unaware they are part of a data collection,' Boult told the Financial Times.
'When you’re watching students on a sidewalk, there’s an awful lot of facing down looking at your phone. In Colorado, where it’s cold and snowy, they cover up in a natural way with scarves and hats. Our goal is to make it the most realistic unconstrained video surveillance facial recognition dataset in the world,' Boult said.
At the time, University of Colorado students had not been informed they were under surveillance nor were they told that images of them would be used to train military and intelligence agency facial recognition systems.
In addition, no infomation was provided as to how they could opt-out or have their photographs removed from the system.
Operator: Beckman Institute; Beihang University; Inception Institute of Artificial Intelligence, Abu Dhabi; Pontificia Universidad Católica de Chile; Queen Mary University of London; University of Notre Dame; Vision Semantics
Developer: University of Colorado
Purpose: Train facial detection and facial recognition systems
Technology: Dataset; Facial recognition; Computer vision
Issue: Privacy; Ethics
Transparency: Governance; Complaints/appeals; Marketing; Privacy
Research, audits, investigations, inquiries, litigation
Harvey, A., LaPlace, J. (2019). Exposing.ai
Murgia M., Financial Times (2019). Who’s using your face? The ugly truth about facial recognition
Cheng Z., Zhu X., Gong S. (2018). Surveillance Face Recognition Challenge
News, commentary, analysis
Published: February 2023