Can off-the-shelf object detectors be used to extract geographic information from geo-referenced social multimedia?

Abstract

On-line photo sharing websites such as Flickr not only allow users to share their precious memories with others, they also act as a repository of all kinds of information carried by their photos and tags. The objective of this work is to perform geographic knowledge discovery by crowdsourcing of geographic information from Flickr's geo-referenced photo collections. In particular, we explore the idea of extracting geographic information semantically for land-use classification by applying state-of-the art object and concept detectors directly to the photo collections. Our results suggest that even though the detectors are able to produce distinctive spatial distributions of different objects, performing land-use classification using user contributed geo-referenced photos remains a challenging problem due to the wide variety of photos available in the collections.