gaze-behavior-annotation-toolkit

# GazeBehavior Annotation Toolkit (GBAT) Methodology from arXiv:2605.22962 (May 2026). Submitted to IEEE ICDL 2026. ## Overview GBAT is a deep-learning-based toolkit designed to automate the annotation pipeline for egocentric eye-tracking and video data of child-caregiver interactions. It addresses three key preprocessing and feature extraction challenges: 1. **Post-hoc synchronization** across multiple video streams 2. **Semi-automatic annotation** of gaze target categories 3. **Categorization** of participants' poses and hand actions ## Key Features - Deep learning models for gaze target detection from egocentric video - Automatic synchronization of multiple camera streams - Pose and hand action classification - Scalable for large-scale and longitudinal developmental studies - Built on modern computer vision architectures ## Applications - Investigating attentional dynamics in naturalistic behavior - Studying how attention interacts with action and language in real time - Large-scale developmental psychology research - Longitudinal studies of early human development ## Trigger Words - gaze behavior, eye tracking, egocentric video, child-caregiver interaction - behavioral annotation, video synchronization, pose estimation - developmental psychology, attention dynamics, naturalistic behavior - GBAT, annotation toolkit ## Related Skills - behavior-vlm-neuroscience