TASIM: Tagged Structures in Mozambique

Published: 16 April 2020| Version 1 | DOI: 10.17632/pr8bgkj3c3.1
Peter Fisker, David Malmgren, Thomas Sohnesen


TASIM (Tagged Structures in Mozambique) is a dataset with tags assigned to all structures based on aerial images. The data sets has 107507 tagged structures in seven categories (house colored roof, house grey roof, house palm roof, non-residential structure, small non-residential structure, structure under construction, apartment). The data cover the entire city of Tete but also include tags from smaller areas in the Mozambique cities Maputo, Beira, Nampula and Quelimane. Each structure’s location and type was tagged by Cloud Factory in Google Earth in September 2016. Subsequently, an image of the location for each structure’s location was downloaded using Google Maps API in November, 2018. The time gap was due to a gap in funding. The use of the images from Google Earth and Google Maps must respect the corresponding terms of use. All images and their associated annotations in TASIM can be used for academic purposes only, but any commercial use is prohibited. The data has been used as part of the paper XX, where is used to train a CNN model prediction structure density and quality. During this work It was noted that the accuracy of the tags re less than ideal. Data files: 1) TASIM.cvs. File includes the variables HH_ID (for merge with images), latitude and longitude, and category. 2) TASIM_Structures.zip. File include images around the location for all tagged structures in 400x400 resolution.



Convolutional Neural Network