Discovering Useful Parts for Pose Estimation in Sparsely Annotated Datasets
Abstract
Our work introduces a novel way to increase pose estimation accuracy by discovering parts from unannotated regions of training images. Discovered parts are used to generate more accurate appearance likelihoods for traditional part-based models like Pictorial Structures and its derivatives. Our experiments on images of a hawkmoth in flight show that our proposed approach significantly improves over existing work for this application, while also being more generally applicable. Our proposed approach localizes landmarks at least twice as accurately as a baseline based on a Mixture of Pictorial Structures (MPS) model. Our unique High-Resolution Moth Flight (HRMF) dataset is made publicly available with annotations.
Cite
Text
Breslav et al. "Discovering Useful Parts for Pose Estimation in Sparsely Annotated Datasets." IEEE/CVF Winter Conference on Applications of Computer Vision, 2016. doi:10.1109/WACV.2016.7477670Markdown
[Breslav et al. "Discovering Useful Parts for Pose Estimation in Sparsely Annotated Datasets." IEEE/CVF Winter Conference on Applications of Computer Vision, 2016.](https://mlanthology.org/wacv/2016/breslav2016wacv-discovering/) doi:10.1109/WACV.2016.7477670BibTeX
@inproceedings{breslav2016wacv-discovering,
title = {{Discovering Useful Parts for Pose Estimation in Sparsely Annotated Datasets}},
author = {Breslav, Mikhail and Hedrick, Tyson L. and Sclaroff, Stan and Betke, Margrit},
booktitle = {IEEE/CVF Winter Conference on Applications of Computer Vision},
year = {2016},
pages = {1-9},
doi = {10.1109/WACV.2016.7477670},
url = {https://mlanthology.org/wacv/2016/breslav2016wacv-discovering/}
}