GBAT AI Toolkit Automates Child-Caregiver Interaction Annotation

Researchers have introduced the GazeBehavior Annotation Toolkit (GBAT), a deep-learning-based system for automating the annotation of egocentric eye-tracking and video recordings of child-caregiver interactions. Manual annotation of such multimodal data is notoriously time-consuming, hindering large-scale studies of how attention, action, and language develop in naturalistic settings. GBAT addresses this by streamlining three critical preprocessing tasks: post-hoc synchronization across multiple video streams, semi-automatic labeling of gaze target categories, and categorization of participants' poses and hand actions.

GBAT leverages modern computer vision and deep learning to extract features at scale, making it possible to conduct longitudinal investigations—tracking the same dyads over weeks or months—that were previously impractical. The toolkit is designed for researchers in developmental psychology, human-computer interaction, and cognitive science. By reducing manual effort and standardizing annotation, GBAT paves the way for more robust, reproducible studies of early social and cognitive development. The team submitted their work to the 2026 IEEE International Conference on Development and Learning.