Copresence: End-to-End 3D Avatar Capture in the Wild
October 21, 2025- Copresence
This talk was presented by me at 3DBODY.TECH 2025.
Abstract
Photorealistic 3D human capture has historically relied on light stages and extensive artist cleanup, restricting deployment to high-budget VFX. Emerging telepresence use cases in VR and AR require at-home capture with commodity devices under uncontrolled conditions. We present Copresence, a mobile pipeline that generates photorealistic 3D avatar assets from a single smartphone scan captured in suboptimal environments. The system integrates (i) an in-app capture workflow that enforces recording quality via guidance and coverage checks, (ii) reconstruction robust to lighting variation, yielding consistent geometry and appearance, and (iii) 360-degree completion from predominantly frontal input, including explicit strand-based hairstyle modeling. We outline design choices, robustness considerations, and practical failure modes, demonstrating a path to high-fidelity avatars without specialized hardware or manual 3D cleanup.
Video
Citation
@inproceedings{leistner2025copresence,
title={Copresence: End-to-End 3D Avatar Capture in the Wild},
author={Leistner, Titus and Grassal, Philip-William and Hormann, Luca and Hamlaoui, Nadia and Ardizzone, Lynton and Iov, Stefan and De Sombre, Philipp and Naef, Sven},
booktitle={Proceedings of 3DBODY.TECH 2025 - 16th International Conference and Expo on 3D/4D Body Scanning, Data and Processing Technologies},
address={Lugano, Switzerland},
note={\#09},
year={2025}
}