Unifying Holistic and Parts-Based Deformable Model Fitting

Joan Alabort-i-Medina, Stefanos Zafeiriou; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 3679-3688

Abstract


The construction and fitting of deformable models that capture the degrees of freedom of articulated objects is one of the most popular areas of research in computer vision. The two main approaches are: Holistic Deformable Models (HDMs), which try to represent the object as a whole, and Parts-Based Deformable Models (PBDMs), which model object parts independently. Both models have their own advantages. In this paper we try to marry the previous two frameworks into a unified one that potentially combines the advantages of both. We do so by merging the popular Active Appearance Models (holistic) and Constrained Local Models (part-based) using a novel probabilistic formulation of the fitting problem. To the best of our knowledge, this is the first time that such an idea has been proposed. We show that our unified holistic and part-based formulation achieves state-of-the-art results in the problem of face alignment in-the-wild. Finally, in order to encourage open research and facilitate future comparisons with the proposed method, our code will be made publicly available to the research community.

Related Material


[pdf]
[bibtex]
@InProceedings{Alabort-i-Medina_2015_CVPR,
author = {Alabort-i-Medina, Joan and Zafeiriou, Stefanos},
title = {Unifying Holistic and Parts-Based Deformable Model Fitting},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2015}
}