Simulation-based modeling of primate social gaze behavior using segmentation-derived visual information

Social gaze behavior provides insight for processing social information across primates, including humans. In this study, we investigated this process in non-human primates by analyzing gaze behavior in rhesus monkeys while they viewed videos depicting social interactions among conspecifics. To quantitatively assess how monkeys interpret socially relevant visual information, we developed a simulation-based framework that models the visual information available in each stimulus frame and compares it with observed viewing patterns. As part of this framework, we developed an automated segmentation algorithm as a tool for extracting face and body regions from dynamic social stimuli. Real-viewing fixation data descriptors were extracted from empirical scanpaths, whereas segmentation-derived descriptors were extracted from manual ground-truth segmentation and the automated segmentation algorithm. Descriptors derived from manual segmentation showed high similarity to real-viewing descriptors, indicating strong alignment with observed gaze behavior. However, descriptors derived from the automated segmentation algorithm also showed strong similarity to real-viewing descriptors and closely followed the manual-segmentation descriptor structure, suggesting that the automated algorithm preserved substantial gaze-relevant visual information. Together, these results demonstrate a quantitative, simulation-driven approach for linking social stimulus features with gaze allocation in non-human primates and provide a methodological foundation that may be extended to human social cognition and translational studies of altered gaze behavior.