Location-based Queries and Query representation for the Lifelog Search Challenge 2021 (Bachelor Thesis, Finished)


Sanja Popovic


Today’s technology allows modern devices to offer high storage capacity that is nowadays more affordable compared to only few years ago. Together with low-cost sensoring devices, all kinds of data accumulate over time which poses a challenge to multimedia retrieval. The annual Lifelog Search Challenge (LSC) is a competition which addresses this problem. The competition gathers multimedia retrieval systems which compete against each other in finding a sought image which is described in words. All systems work on the same dataset which stems from a lifelogger who continuously took images over multiple months. One of the systems that participates each year in the LSC is vitrivr, a multi-modal retrieval system. Since image descriptions in LSC tasks often include spatial information, the idea was to extend vitrivr by a new modality, i.e., location-based search. In this thesis, vitrivr is extended by a new location-based query formulation and presentation. The thesis also discusses the applied concepts and evaluates the user experience which finally led to encouraging results.



Start / End Dates

2021/03/29 - 2021/07/28


Research Topics