Parametric Skeleton Generation via Gaussian Mixture Models
Abstract
We propose an efficient and effective control point extraction algorithm for parametric skeleton generation. The object skeleton pixels are predicted via an hourglass network and partitioned into skeleton branches using Gaussian Mixture Models. For each skeleton branch, a Bezier curve is utilized to generate the control points. The radius of the skeleton is computed by the distance between the border of the object and the Bezier curve. The branches are sorted by the area so that the parametric skeleton representation is unique. For the Parametric SkelNetOn competition, the proposed approach achieves the prediction score of 11793.89, which is in the first place on the performance leader-board.
Cite
Text
Liu et al. "Parametric Skeleton Generation via Gaussian Mixture Models." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019. doi:10.1109/CVPRW.2019.00153Markdown
[Liu et al. "Parametric Skeleton Generation via Gaussian Mixture Models." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019.](https://mlanthology.org/cvprw/2019/liu2019cvprw-parametric/) doi:10.1109/CVPRW.2019.00153BibTeX
@inproceedings{liu2019cvprw-parametric,
title = {{Parametric Skeleton Generation via Gaussian Mixture Models}},
author = {Liu, Chang and Luo, Dezhao and Zhang, Yifei and Ke, Wei and Wan, Fang and Ye, Qixiang},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year = {2019},
pages = {1167-1171},
doi = {10.1109/CVPRW.2019.00153},
url = {https://mlanthology.org/cvprw/2019/liu2019cvprw-parametric/}
}