Semantic 3D Reconstruction of Heads
Abstract
We present a novel approach that jointly reconstructs the geometry of a human head and semantically segments it into labels such as skin, hair and eyebrows. In order to get faithful reconstructions from data captured in uncontrolled environments, we propose to adapt a recently introduced implicit volumetric surface normal based shape prior formulation. Shape prior based approaches critically rely on an accurate alignment between the data and the prior to succeed. To this end, we propose an automatic alignment procedure for the used shape prior formulation. We evaluate our alignment procedure thoroughly and show head reconstruction results on challenging datasets.
Cite
Text
Maninchedda et al. "Semantic 3D Reconstruction of Heads." European Conference on Computer Vision, 2016. doi:10.1007/978-3-319-46466-4_40Markdown
[Maninchedda et al. "Semantic 3D Reconstruction of Heads." European Conference on Computer Vision, 2016.](https://mlanthology.org/eccv/2016/maninchedda2016eccv-semantic/) doi:10.1007/978-3-319-46466-4_40BibTeX
@inproceedings{maninchedda2016eccv-semantic,
title = {{Semantic 3D Reconstruction of Heads}},
author = {Maninchedda, Fabio and Häne, Christian and Jacquet, Bastien and Delaunoy, Amaël and Pollefeys, Marc},
booktitle = {European Conference on Computer Vision},
year = {2016},
pages = {667-683},
doi = {10.1007/978-3-319-46466-4_40},
url = {https://mlanthology.org/eccv/2016/maninchedda2016eccv-semantic/}
}