Date Available
8-13-2023
Year of Publication
2022
Degree Name
Doctor of Philosophy (PhD)
Document Type
Doctoral Dissertation
College
Engineering
Department/School/Program
Computer Science
First Advisor
Dr. Nathan Jacobs
Abstract
The problem of estimating the location from which un-geotagged photographs were captured has been well studied by the computer vision community in recent years. The central proposal of this thesis is to define a common framework within which existing approaches can be constructed and evaluated, and to introduce a new method under this framework which uses cross-attention between the query image and a database of satellite imagery with known geotags. Our experiments fit within three broad categories: 1) evaluating the ability of image localization approaches to generalize to unseen regions; 2) examining performance changes under various reference database resolutions, scales, and densities; and 3) exploring localization with multi-modal reference imagery. Our key contribution is the notion of attending between query and reference imagery throughout inference, compared with the existing practices of attending late or not at all.
Digital Object Identifier (DOI)
https://doi.org/10.13023/etd.2022.341
Funding Information
Portions of this work was supported by the Intelligence Advanced Research Projects Activity (IARPA) via contract 2017-16110300001, in 2021.
Recommended Citation
Greenwell, Connor, "Image Geo-localization with Cross-Attention" (2022). Theses and Dissertations--Computer Science. 120.
https://uknowledge.uky.edu/cs_etds/120
