Date: June 2019.
Source: 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10934-10943.
Research Summary: A 3D morphable model, or 3DMM, is a statistical model that can generate or reconstruct 3D shapes across a population, built from many real scans. Different research teams have often built separate 3DMMs for different parts of the same object, for instance one detailed model of the face, and a separate model covering the full head and cranium, without an established way to combine them. This paper proposes two methods for merging two such models together, even when they were built from different, non-overlapping datasets. As a working example, the authors combine a detailed face model with a full head and cranium model, producing a single, more capable model that outperforms the original cranium model on every measure tested.
3dMD’s Role: This model combines two entirely separate 3dMD datasets, from two independent 3dMD customer relationships. The facial detail comes from MeIn3D, roughly 10,000 scans captured at Imperial College London’s 2012 “Me in 3D” event. The cranial detail comes from Headspace, 1,519 subjects captured on a static 5-view 3dMDhead system by the Alder Hey Children’s Hospital Craniofacial Unit in Liverpool, led by Christian Duncan. Two independently captured, real-world 3dMD datasets, gathered years apart by different institutions for different purposes, combine here into a single, more capable model.
Article: Combining 3D Morphable Models: A Large Scale Face-and-Head Model.
Authors: Stylianos Ploumpis, Haoyang Wang, Nick Pears, William A. P. Smith, Stefanos Zafeiriou. Imperial College London, University of York, FaceSoft.io