Data Cities Meetup: Behind Facial Recognition Datasets
Disruption Network Lab
Following on the 20th conference DATA CITIES: Smart Technologies, Tracking & Human Rights (25-27 September 2020), Disruption Network Lab invite to join an evening with artist & researcher Adam Harvey, to explore facial recognition datasets being made in cities across the world, and understand more about where this data originates and who is using it.
Have you ever been recorded by a surveillance camera? Posted a selfie online? Shared a photo album on Flickr? If you answered yes, it is very likely that your images are being used for training computer vision surveillance algorithms. Artificial Intelligence has grown out of the massive amounts of data being created by our increasingly digitized lives. But where does that data originate and who is using it?
MegaPixels is a research project by Adam Harvey and Jules LaPlace that tracks, analyzes, and publishes about the origins and endpoints of image training datasets. Join us tonight to explore datasets being made in cities across the world. Maybe you’ll even find yourself.
Adam Harvey is an artist and researcher focused on privacy, surveillance, and computer vision. His past projects include developing camouflage techniques for subverting face detection, thermal imaging, and location tracking. He is the founder of vframe.io, a computer vision project for human rights researchers; and is currently a researcher-in-residence at Karlsruhe HfG. His most recent project MegaPixels explores the image datasets and information supply chains contributing to the growing crisis of authoritarian biometric surveillance technologies.
The MegaPixels project first launched in 2017 for an installation at Tactical Technology Collective’s GlassRoom about face recognition datasets. Then a facial recognition kiosk, the project expanded in 2018 with a commission by Elevate Arts festival in Austria, as part of Re-Imagine Europe, to analyze pedestrian datasets. Since then this project has evolved from an analysis of 3 datasets to large-scale interrogation of hundreds of image training datasets.