Abstract
Research across face recognition, social cognition, and clinical psychology treats visual experience as a primary mechanism for learning and successful categorization. As the field has shifted toward naturalistic, dynamic stimuli, there is a need to dissociate between exposure, which is the amount of time a target is presented, and experience, which is the amount of time an observer is looking at a certain target. Here, we present a pipeline that dissociates exposure from experience with faces during naturalistic viewing of a movie with multiple characters. This is achieved by combining screen-based eye tracking with deep neural network-based face detection and recognition to measure each participant's gaze-based visual experience with faces during naturalistic viewing, on a frame-by-frame basis. With a sample stimulus, we benchmarked our pipeline as an identity exposure extractor and achieved high face detection (~98%) and face recognition (85-88%) accuracies. In a proof-of-concept experiment, participants watched two episodes of a television series while their gaze was recorded. Our pipeline revealed participants looked at individual characters for less than 27% of their appearance time on average indicating that character appearance time overestimated participants' actual visual experience. Moreover, participants showed reliable individual differences in face-directed gaze, both in overall face-viewing tendencies and in character-specific preferences, with stable idiosyncratic preferences emerging when two characters competed for visual attention. Together, these findings demonstrate a meaningful gap between exposure and idiosyncratic gaze-based visual experience, highlighting the importance of dissociating experience from exposure when studying the role of visual experience in naturalistic viewing.