68 Landmarks started life as a broadsheet zine and an extension of my degree show project and academic dissertation research findings. 68 Landmarks explores ideas around the inherent bias within the data sets of facial detection and tracking algorithms and the questionable methods involved in training these neural networks. Training involves using repositories of images of faces, often of western celebrities, public figures and icons because of the vast number of images available of these types of individuals. These images are then used in the training of deep neural networks through the manual marking of the borders of facial features on a large volume of images of faces in a variety of expressions.
These 68 facial marks comprise an Active Shape Model, enabling the algorithm to identify each face in the training set. In my work, I use the Active Shape Model as a mask, overlaying and slightly obscuring press images of celebrities and public figures. This acts as a signifier of a tribe of people training our facial tracking and detection software, much like African tribal face markings and masks. I recently launched 68landmarks.com to accompany the publication, and this website accesses the user’s webcam and overlays the 68 Landmarks using facial detection and AR. It also allows users to capture, download and share their own unique digitally manipulated portraits, which are playful but have a significant underlying message.