Abstract
Observers free-viewing scenes reliably vary along oculomotor metrics as well as their tendency to fixate objects with specific semantic attributes, like faces and text. Other findings suggest some of these gaze biases are heritable and associated with neurodevelopmental disorders. However, most studies examining free gaze behaviour used small samples, unlikely to reflect the breadth of human variability. Here, we moved towards a more comprehensive understanding of individual gaze by collecting a massive dataset of scene viewing behaviour outside the lab. We collaborated with a science museum with 150k visitors per year to build a fully automated eye-tracking exhibit. During participation, subjects pass through a survey and calibration routine before freely viewing a video and 40 natural scenes - a short test of individual gaze, which we previously validated. So far, we collected > 2 million fixations from > 6,000 female (54%), male (42%) and diverse (4%) observers, ranging from 4-85 years of age. We find large variability for oculomotor measures like avg. fixation duration (3-fold range: 239ms-771ms) and saccadic frequency (4-fold range: 0.7-3.3 Hz), as well as for individual salience biases, like the central-bias (4-fold range), the proportion of dwell time on text (0% - 53%) and faces (9-fold range: 9% - 86%). The large and diverse nature of our sample allows us to tackle novel questions regarding the development and diagnostic potential of individual gaze. A first analysis revealed surprisingly late plasticity, showing a 2-fold, linear increase in text salience from age 4 to 24 (R^2 = 93% for age-wise bins), which was complemented by a 14% decrease in face salience; as well as a U-shaped central-bias development across the lifespan. In ongoing clinical collaborations, we collect data from children with ASD and ADHD, allowing us to probe the diagnostic potential of our paradigm against a massive benchmark sample.