Spatial Relationship-Driven Computer Vision Image Data Set Annotation
Jeremy Davis, James B. Haynie, Derek T. Anderson, Cindy L. Bethel, J. Edward Swan II, John E. Ball, and Amy Bednar. Spatial Relationship-Driven Computer Vision Image Data Set Annotation. In IEEE International Joint Conference on Neural Networks (IJCNN), IEEE, July 2022. DOI: 10.1109/IJCNN55064.2022.9892975.
Download
Abstract
Modern machine learning (ML) is based to a great extent on supervised deep learning models that require large amounts of labeled training data. While image data sets with annotations exist, the annotations are produced manually and possess relatively simple descriptions. To date, none of the freely available labeled image data sets incorporate spatial reasoning, one of Gardner's nine human intelligences. This article presents a new process with open source tools provided to label imagery based on spatial interactions between image objects and automated reasoning under uncertainty. The resulting annotated data can be used to train new ML/AI algorithms and/or help us better understand existing methodologies.
BibTeX
@InProceedings{IJCNN22,
author = {Jeremy Davis and James B. Haynie and Derek T. Anderson and
Cindy L. Bethel and J. Edward {Swan~II} and John E. Ball
and Amy Bednar},
title = {Spatial Relationship-Driven Computer Vision Image Data Set
Annotation},
booktitle = {IEEE International Joint Conference on Neural Networks (IJCNN)},
year = 2022,
location = {Padova, Italy},
publisher = {IEEE},
date = {July 18--23},
month = {July},
note = {DOI: <a target="_blank"
href="https://doi.org/10.1109/IJCNN55064.2022.9892975">
10.1109/IJCNN55064.2022.9892975</a>.}
abstract = {
Modern machine learning (ML) is based to a great extent on supervised deep
learning models that require large amounts of labeled training data. While image
data sets with annotations exist, the annotations are produced manually and
possess relatively simple descriptions. To date, none of the freely available
labeled image data sets incorporate spatial reasoning, one of Gardner's nine
human intelligences. This article presents a new process with open source tools
provided to label imagery based on spatial interactions between image objects
and automated reasoning under uncertainty. The resulting annotated data can be
used to train new ML/AI algorithms and/or help us better understand existing
methodologies.
},
}