Extending Convolutional Pose Machines for Facial Landmark Localization in 3D Point Clouds

Eimear O' Sullivan; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

Abstract


In this work we address the problem of landmark localization in 3D point clouds by extending the convolutional pose machine (CPM) architecture to facilitate landmark localization in 3D point clouds. Making use of PointNet++,we are able to construct an architecture that is invariant tothe ordering of an input point cloud. The sequential CPM architecture facilitates allows initial heatmaps to be iteratively refined in a series of point convolutional stages to yield robust landmark predictions. We propose to evaluate our approach for 3D facial landmark localization on benchmark face databases, BU-3DFE, BP4D-Spontaneous and BP4D+. The robustness of the approach to the size of the input point cloud will be assessed, and the contribution of the CPM stages will be evaluated in an ablation study.

Related Material


[pdf]
[bibtex]
@InProceedings{Sullivan_2019_ICCV,
author = {O' Sullivan, Eimear},
title = {Extending Convolutional Pose Machines for Facial Landmark Localization in 3D Point Clouds},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2019}
}