Date: April 2017.
Source: International Journal of Computer Vision, 126:233–254
https://doi.org/10.1007/s11263-017-1009-7.
Research Summary: Large Scale Facial Model (LSFM) is a 3D Morphable Model (3DMM) automatically constructed from 9,663 distinct facial identities, making it, at the time of publication, the largest-scale morphable model ever built, capturing statistical information across a wide variety of the human population. The paper introduces a fully automated, robust pipeline for constructing such a model, informed by an evaluation of state-of-the-art dense correspondence techniques. Because the underlying dataset includes rich demographic information for each subject, the pipeline supports building not only a single global 3DMM but also separate models tailored to specific age, gender, or ethnicity groups. The authors use LSFM to perform age classification from 3D shape alone and to reconstruct noisy out-of-sample data within the model’s low-dimensional space, and extensive qualitative and quantitative evaluation shows the resulting 3DMM outperforms existing models by a large margin. The team makes the automatic construction pipeline’s source code publicly available, along with the global model and its age, gender, and ethnicity-specific variants.
3dMD’s Role: LSFM is trained on real-world scans from MeIn3D, a database of approximately 12,000 3D facial scans captured at the Science Museum, London’s “Me in 3D” event, part of the museum’s Live Science programme, running from 11 January to 10 April, 2012, on a static 3dMDtrio system. LSFM itself uses 9,663 of MeIn3D’s scans after quality filtering.
Article: Large Scale 3D Morphable Models.
Authors: James Booth, Anastasios Roussos, Allan Ponniah, David Dunaway, Stefanos Zafeiriou. Imperial College London and Great Ormond Street Hospital, London, UK.