Image and Video Analysis

European Cities 1M dataset

European Cities 1M dataset consists of 909,940 geo-tagged images from 22 European cities, crawled from Flickr using geographic queries covering a window of each city center. A subset of 1,081 images from Barcelona is annotated into 35 groups depicting the same scene; 17 of the groups are landmark scenes and 18 are non-landmark. Since not all scenes are landmarks, annotation cannot rely on tags; it is rather a combination of visual query expansion and manual clean-up. Five images are selected as queries from each group. If the group contains less than 5 images, all of them are used as a query (which is frequent for non-landmark scenes). In total, we used 157 queries.

We have not included any other photos from Barcelona in the dataset, to ensure no other image depicts the same scene/building as the ground truth. The remaining 908,859 images from the other 21 cities are the distractors. Most of them depict urban scenery like the ground-truth, making a challenging distractor dataset.

Contrary to our European Cities 50K dataset, where the query images come from 9 European cities, here we chose all query images from Barcelona. This is because, for this dataset, we need to maximize the number of distractor images. For each city with query images we exclude from the distractor dataset all other images except for the annotated ones. Therefore we pick queries from just one city.

Sets of query images selected for evaluation for Landmark and Non-landmark scenes are depicted in Figures 1 and 2 respectively, while sample images from the distractor set are presented in Figure 3.

Figure 1. Query images, one for each of the 17 landmark scenes in the annotated EC1M dataset.

Figure 2. Query images, one for each of the 18 non-landmark scenes in the annotated EC1M dataset.

Figure 3. Sample distractor images from the EC1M dataset.

Dataset Rights

The EC1M dataset consists of images downloaded from Flickr. Use of these images must respect the corresponding terms of use.

Citation

If you use the EC1M dataset please cite our ACM Multimedia paper shown below.

Downloads

The annotated set of 1081 images is provided for downloading in two compressed files: the one contains the landmark scenes and the other the non-landmark scenes. Images are in JPEG format with a maximum resolution of 500x500 pixels. The 157 queries are a subset of the annotated set and their filenames are found in the corresponding text file below. The set of 908K images used as distractors can be downloaded from Flickr using the direct links provided in the compressed text file below. Frequently, some of the images are moved amongst Flickr servers. We periodically update the list of URLs replacing broken links with the current ones.

Annotated set of landmark scene images (927 images, 117.5 MB)
Annotated set of non-landmark scene images (154 images, 21.4 MB)
Queries (1.3 KB)
Distractor images (908K images, 14.0 MB) [ Last update 9 April 2011 ]

NEW! Annotated set of landmark scene images geotags (927 images, 3 KB)
NEW! Annotated set of non-landmark scene images geotags (154 images, 1 KB)
NEW! Distractor images geotags (908K images, 3.0 MB)

Publications

Conferences

Y. Avrithis, Y. Kalantidis, G. Tolias, E. Spyrou. Retrieving Landmark and Non-Landmark Images from Community Photo Collections. In Proceedings of ACM Multimedia (MM 2010), Firenze, Italy, October 2010.
[ Abstract ]
[ Bibtex ] [ PDF ]

Journals

Y. Kalantidis, G. Tolias, Y. Avrithis, M. Phinikettos, E. Spyrou, P. Mylonas, S. Kollias. VIRaL: Visual Image Retrieval and Localization. In Multimedia Tools and Applications (to appear), 2011.
[ Abstract ]
[ Bibtex ] [ PDF ]