68 Landmarks refers to the standard number of points that make up an ‘Active Shape Model’, used to detect and track faces in facial detection and tracking software.
An Active Shape Model is an algorithmic statistical model of a face shape. These deep neural networks have been trained by people manually marking the borders of facial features on a vast number of images of faces in a variety of expressions. Over time, as the network is trained on more faces, the algorithm becomes more efficient at accurately mapping the face, and the network outputs a vector for each face. This is known as an Active Shape Model enabling the algorithm to identify each face in the training set.
Bias has long plagued facial detection algorithms. The problem occurs with subjects in the images that neural networks are trained on, as well as the companies and individuals developing the software itself. The software is often developed in the West by non-ethnically diverse teams. They use large repositories of images to train the neural networks, often primarily containing images of western celebrities, public figures and icons. Because there are a high number of images available of these kinds of individuals online, uploaded to social media and taken by paparazzi, they can be easily downloaded into a database. The problem is that these individuals are often wealthy Caucasian, with very symmetrical, airbrushed and potentially medically enhanced faces with little to no visible impurities. Therefore, training neural networks on these kinds of images creates a bias, because the algorithm does not represent the wider community of people that will inevitably interact with and ultimately be affected by these technologies which are becoming a more prominent feature of their daily lives.
In 2019 produced a newsprint publication to give the project more detailed contextualization. The publication documents a body of work, which much like this website aims to highlight the problematic nature of training facial detection and tracking algorithms and the inherent bias ingrained within it. I produced a series of eight artworks, using face-tracking software and targeting well-known celebrities and public figures, whose image has been captured reproduced and distributed many times. Through the work, I aim to raise the question, who are we ultimately serving through the development of our modern technologies?
Some of the public figures featured in the publication are the most photographed and well-documented individuals ever with undeniable lasting legacies, however, their ultimate legacy could be less obvious. The permanent mark they leave inside the neural networks, which enable us to use our face to seamlessly walk through an airport barrier, unlock our phones or instantly authenticate a bank transfer.
This website was made to accompany the small run of newspapaer zines, it is also a continuation, to extend the discussion beyond the publication. The website is interactive, and allows the user to create, download and share their own unique digitally manipulated augmented reality portraits, which are playful but will with a significant underlying message.
68landmarks.com was created by Radley Cook
using the Beyond Reality Face
SDK by Tastenkunst
| © Radley Cook 2019