Geolocation Estimation - Tool for geolocation estimation of photos

URL: https://labs.tib.eu/geoestimation

This tool demonstrates the automatic geolocation estimation of photos using deep neural networks. In addition to analyzing their own photos, users can compete against the artificial intelligence in a quiz.

To train the deep learning models, approximately five million images from Flickr with Creative Commons licenses were used. For this purpose, the Earth was divided into geographical sectors with different geographical resolutions, and a classification approach was trained to estimate the correct sector for an input image and convert it into a GPS coordinate. To learn relevant features for urban, natural and indoor environments, individual models were trained with images from these particular environments. The approach was presented at the European Conference on Computer Vision 2018 and was able to outperform comparable approaches from Google research groups, which used significantly more training images, on standard benchmarks.

You can find the details in the following publication.

E. Müller-Budack, K. Pustu-Iren, and R. Ewerth:
Geolocation Estimation of Photos Using a Hierarchical Model and Scene Classification
In: European Conference on Computer Vision, ECCV 2018, Munich, Germany, September 8-14, 2018, pp. 575–592. Springer, 2018.
https://doi.org/10.1007/978-3-030-01258-8_35

Images

Screenshot of the geolocation estimation tool showing the results for a building located in Boston, Massachusetts, USA