Landmark Mining and Recognition based on Community Photo Collections
Tobias Weyand and Bastian Leibe
Community photo collections have become a valuable source for large amounts of tourist photos, densely covering entire cities. In particular, they provide rich imagery of the world's landmark buildings, statues, monuments, and paintings. Our goal is to automatically discover popular objects in community photo collections by clustering the images by the depicted object and to find a representative iconic view for each of them. These clusters can then be used to automatically construct landmark recognition systems for use in mobile visual search apps or automatic annotation of holiday photos. Other applications include scene summarization and 3D building reconstruction for use in virtual city models.
Discovering Details and Scene Structure with Hierarchical Iconoid Shift, T. Weyand, B. Leibe in International Conference on Computer Vision (ICCV'13), 2013. (PDF)
Discovering Favorite Views of Popular Places with Iconoid Shift, T. Weyand, B. Leibe in International Conference on Computer Vision (ICCV'11), 2011 (oral presentation). (PDF))
An Evaluation of Two Automatic Landmark Building Discovery Algorithms for City Reconstruction, T. Weyand, J. Hosang, B. Leibe in ECCV'10 Workshop on Reconstruction and Modeling of Large-Scale 3D Virtual Environments (RMLE'10), 2010. (PDF)
500k geotagged images of the inner city of Paris, collected from Flickr and Panoramio (Dataset). The dataset was used for the evaluation of large-scale landmark discovery algorithms in An Evaluation of Two Automatic Landmark Building Discovery Algorithms for City Reconstruction (PDF)
For additional information please contact Tobias Weyand