Revision of ImageCLEF 2008: Visual Concept Detection Task from Mon, 03/10/2008 - 10:55

In 2008, ImageCLEF will offer a Visual Concept Detection Task continuing the Object Annotation Task of ImageCLEF 2006 and the Object Retrieval Task of ImageCLEF 2007. In contrast to its predecessors, the Visual Concept Detection Task of ImageCLEF 2008 will be strongly interacting with the ImageCLEF 2008 Photographic Retrieval Task.

Outline

The Visual Concept Detection Task has the objective to identify visual concepts that would help in solving the photographic retrieval task in ImageCLEF 2008. Therefore, we will publish a training database of approximately 2500 images with annotated visual concepts. Then, the objective for participating groups is to identify these visual concepts in the 20,000 images that are used for the retrieval task.

At a later stage, the usefullness of the detected topics for the retrieval procedure will be evaluated.

Data

The visual concept detection task will work on a subset of the extended IAPR TC-12 database which is also used for the Photo Retrieval Task of ImageCLEF 2008.

We will be using approx 1500 images for training and 1000 images for testing, and maybe 20,000 images as extended test set. If the extended test set is used, the performance of the annotation will not directly be assessed but the usefulness of the annotation for the retrieval task will be assessed.

Schedule

The schedule is not yet defined.

We are trying to fit the schedule BEFORE the Photo Retrieval Task of ImageCLEF 2008 so as to allow the participants of the photo retrieval task to use the outcome of the visual concept detection, but we are still in the process of annotation the training data.

Organisers

Thomas Deselaers, RWTH Aachen University, Aachen, Germany
Allan Hanbury, Vienna University of Technology Institute of Computer Aided Automation, Pattern Recognition and Image Processing Group, Vienna, Austria