Format of results for near-duplicate detection image trackResults should be returned as one or several XML files. Each XML file contains answers for one or several tasks.
XML must have the following structure:
<?xml version="1.0"?> <romip:taskresult xmlns:romip="http://www.romip.ru/data/classification" collectionId="ROMIP_IMG-YANDEX" trackId="img-duplicate-2008">Tag run is descriprion of the system produced the result.
Tag cluster describes group of near duplicates found by participant. Every group consists of images that are treated as near-duplicates to each other: every pair of images from the group is treated as near duplicates.
To identify an image (attribute docID of document tag) correspondent identifiers from the test collection should be used.