Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Variability in Young Children's Everyday Visual Experiences of Object Categories

Jane Yang¹, Tarun Sepuri¹, Alvin Wei Ming Tan², Khai Loong Aw², Michael Frank², Bria Lorelle Long¹; ¹University of California, San Diego, ²Stanford University

Presenter: Jane Yang

Children learn to categorize objects from everyday visual experience. How variable are the exemplars and views of objects in that experience – for example, of birds, cribs, and cups? Here, we quantified the variability of the object categories detected in videos taken from the child’s perspective in the BabyView dataset (Long et al., 2024). Detected objects incorporate both viewpoint variability (the range of angles and scales at which an object is seen) and exemplar variability (the range of distinct instances within a category). Using embeddings from both a vision–language model (CLIP; Radford et al., 2021) and a self-supervised vision model (DINOv3; Simeoni et al., 2025), we quantified global dispersion (mean distance from each exemplar to its within-category centroid) and local dispersion (mean within-category k-nearest-neighbor distance, k = 5) on the same set of 7,018 human-validated crops spanning 85 CDI noun categories. Both metrics varied widely across categories, with moderate model agreement moderate for global dispersion (𝜌 = .55) and local dispersion (𝜌 = .63). Dispersion metrics were weakly associated with how frequently a category was detected in egocentric video: categories could appear equally often but differ in visual variability. Our results motivate future work that examines how category frequency and variability interact to yield robust category representations during everyday learning in early childhood.

Topic Area: Development, Individual Differences & Clinical Populations