Fusion Task

Short description

Goal of the fusion task is to explore strategies for fusing data of visual and textual retrieval systems for an optimal multi-modal retrieval performance. We plan to make training data available, so data on a very similar database and also the ground truth. All data will be in the trec_eval form so use of trec_eval is necessary for the testing. trec_eval is available from here.
Registration for the fusion task is free of charge and can be done here.

Please keep in mind that this task does not require image processing as no data base will be distributed, only the results of textual and visual information retrieval systems.

Contact person

• Jayashre Kaplathy-Cramer, OHSU, Portland, OR, USA, kalpathy@ohsu.edu
• Henning Müller, University of Applied Sciences Western Switzerland, Sierre, henning.mueller@sim.hcuge.ch

Important dates

15.11.2009 Release of training data (released 16.11.)
30.11.2009 Release of test data (released 30.11.)
4.1.2010 Submission of results of the participants
10.1.2010 Release of ground truth data
15.1.2010 Paper submission for the main conference
(31.3.2010 Submission of results with ground truth data, as comparison with other results)
30.04.2010 Deadline for overview papers for contests (written by us, the contest organizers)
23.8.-26.8.2010 ICPR Conference

Data sets

Interested participants will receive the best four purely visual and the four best purely textual runs of the 2009 ImageCLEF medical retrieval task and then need to combine the runs in a way to optimize the performance.

Format of the supplied runs and of the results to submit is the trec_eval format.

No images are necessary, and so no image databases is distributed.

Performance measures

All performance measures of trec_eval will be made available.

Lead measure will be mean average precision (MAP) but early precision (P@10, P@30) will also be used for rankings.

How to register for the task

The online registration system is ready here.