Carl Bunce¹ (c.bunce@reading.ac.uk), Katie L. H. Gray¹, Peter Scarfe¹; ¹University of Reading

Research on face and body perception has benefited enormously from large, well-curated stimulus databases. However, nearly all publicly available sets consist of 2D photographs or videos, which is becoming increasingly at odds with the growing use of immersive virtual reality (VR) in vision science. VR offers many advantages for social vision research over traditional screen-based methods, including naturalistic depth cues, full-field visual presentation, a strong sense of presence and immersion, and access to rich behavioural measures—such as precise tracking of eye gaze and head and body position—that can be used to reliably infer implicit social and perceptual processes. Yet progress in this area has been slowed by a lack of realistic-looking 3D human stimuli suitable to present in such tasks. Here we introduce the Solid Sight Social Vision Database, a collection of highly realistic full-body human avatars designed for real-time rendering in VR. The database was created by combining state-of-the-art modelling techniques to achieve high geometric accuracy and photorealistic textures of real-life individuals. Participant likenesses were captured using i) a structured-light 3D scanner (Einscan Pro HD), providing sub-millimeter scanning accuracy, and ii) photogrammetric reconstruction from multi-angle photographs. Raw scan data underwent a post-processing workflow involving alignment, geometry and texture cleaning, and scale and position standardization. We additionally captured multiple facial expressions per participant, producing separate meshes for different emotions. The resulting meshes from this pipeline form a high-quality avatar set intended to support VR-based social vision research, as well as any other research applications that require realistic human models for 2D or 3D rendering outside of VR. The database will be made freely accessible to researchers alongside sample MATLAB and Psychtoolbox code to simplify integration into experimental tasks, including immersive VR.

Acknowledgements: This project was supported by an award from the Leverhulme Trust.