Stubborn Vectors is an interactive installation that explores the computational gaze of computer vision systems, which often oversimplify the complexities of our corporeal selves.
Computer vision is deeply entwined with surveillance technologies, accelerating the extraction of information to classify, analyse, and predict human behaviours—often capturing far more than individuals intend to reveal. This work focuses on Affection Mapping, a branch of machine learning dedicated to classifying emotions using computer vision. However, emotions are complex, evolving in response to our families, environments, and personal histories. This inherent complexity exposes the limitations of models attempting to construct a rigid taxonomy of human emotion.
In their pursuit to classify and represent the physical world, these systems inadvertently reveal an unfamiliar umwelt—a new lens through which meaning is extracted and reality is interpreted. By examining this computational gaze, we can observe how these systems function, their inherent biases, and how they influence and bleed into our physical world.
The installation is built as a computational system composed of computer vision and machine learning models. A dataset of archived footage is processed through this system to detect, segment, and extract features from each frame, with each feature represented as a multi-dimensional vector or embedding. This data is then sorted, reduced in dimensionality, and mapped into a two-dimensional space, creating a human-readable representation of the system's latent space. Using custom-built controllers, users can query the system with an emotional state, prompting it to navigate toward its nearest embedding within the space.
