A multimodal dataset for reconstructing common marmoset body-environment interactions in a 3D digital-twin framework

The common marmoset (Callithrix jacchus) is an important non-human primate model in neuroscience and biomedical research. However, existing 3D resources for this species have mainly focused on brain atlases or keypoint-based pose estimation, and reusable data resources that jointly describe the body surface, fur, articulated structure, and experimental environment remain limited. Here, we present a multimodal dataset designed to reconstruct body- environment interactions of common marmosets in three dimensions. The dataset includes a whole-body surface mesh derived from computed tomography (CT) images, fur representations based on photographic references, a rigged 3D model for pose-driven animation, synchronized behavioral videos from three individuals recorded for approximately 90 hours from eight view-points, 2D and 3D keypoint estimation data, 3D models of the experimental environment constructed from blueprint information, and rendered pseudo-egocentric views generated by integrating pose estimation results with the 3D body and environment models. Technical validation assessed the geometric agreement between the CT-derived mesh and the surface model, the accuracy of 2D and 3D keypoint estimation, the dimensional accuracy of the environment model, and the structural similarity between real and rendered images. This dataset provides a foundation for treating marmoset natural behavior not only as point trajectories but also as a three-dimensional phenomenon involving body shape and its spatial relationship with the environment, thereby enabling applications in behavioral analysis, visualization, synthetic-data generation, and future digital-twin studies.