This document describes the contents and format of Forum 1 dataset. === I. Contents of the dataset === We distribute the original images from Flickr, as well as the associated set of SIFT key files. The provided files are as follows. list.txt -- the list of image files gt_graph.txt -- the ground truth graph keys.tar.gz -- the SIFT key files images.tar.gz -- the image files === II. Description of file formats === 1. List file Each line in the list file specifies the filename to images in jpg format. In addition, the focal length of the image has been estimated from Exif tags is also included. There are three fields for each line: the image name, a zero, and the Exif focal length. The second field is always zero but may change in future datasets. Example: images/3657786028_2193f81c2f_o.jpg 0 1305.05710 For this image, 3657786028 is the Flickr photo id. You can access the original Flickr photo through the Web with a URL of the form: http://www.flickr.com/photo_zoom.gne?id= 2. Ground truth graph file The ground truth graph file contains all geometrically consistent image pairs. Each line is a pair of images, specified by their Flickr photo id. Example: 100589560 438689211 This means image 100589560 and image 438689211 are geometrically consitent and they pass the SIFT matching. 3. Key files We provde the (gzipped) SIFT key files. The format of the key files if the same as that produced by David Lowe's SIFT extractor, described here: http://www.cs.ubc.ca/~lowe/keypoints/ 4. Image files We provide the original jpg file in a .tar.gz file. The following command can be used to extract the content from the file. tar -zxvf images.tar.gz === III. Related publication === If you use this dataset in a publication, please cite the paper: Yin Lou, Noah Snavely, and Johannes Gehrke. "MatchMiner: Efficient Spanning Structure Mining in Large Image Collections." In Proceedings of the 12th European Conference on Computer Vision (ECCV), Florence, Italy, 2012. === IV. Questions === If you have questions with the dataset, please email yinlou@cs.cornell.edu.