Task

Create a repositiory for shared code

Document a standard access protocol

Define criteria for indexing the images

Build tools that help indexing images

Write a tool that creates the ground truth data structure

Test the set-up with existing benchmarks

Publicize the event

Manage images and metadata

Create a repositiory for shared code ^

We are using About all the information given there can be changed. The perl license is a disjunction of the LGPL and the Artistic License. As long as there is no code submitted, changes should be possible (as said, it has to be sorted out with people at SourceForge). All the information you find about the project can be changed, you just need to suggest something if you are not happy with the current version Further instructions

Document a standard access protocol ^

The old (February 2001) images are here We are currently rebuilding the image repository following the directory structure used by the University of California at Berkeley Digital Library Project; we hope this will allow us to piggy-back on their research, described here The physical location of the images is http://www.benchathlon.net/img/done/ and the next task is to install a database to capture the metadata. We can then use Berkeley's cgi script to present a brower interface to the images, which can be used for the indexing Instructions

	Define criteria for indexing the images ^
	Based on the MPEG templates, structured annotation types, and structured annotation DS The results will be a description of the procedure and A simple database containing this metadata

	Build tools that help indexing images ^
	Invent an annotation tool. annotate is currently in an advanced alpha stage, simple help is available Using the production annotation tool, present an image and the possible indexing terms. For each image at least three people will determine the indexing terms. This data is pooled Instructions

	Write a tool that creates the ground truth data structure ^
	Write a tool that creates the ground truth data structure as proposed in the EI 2001 paper, or a better system

	Test the set-up with existing benchmarks ^
	Test the set-up with existing benchmarks, such as BIRDS-I and the Viper benchmark described at http://viper.unige.ch/~muellerh/benchmark2001.ps (if you have a slow link, download this smaller PDF version)

	Publicize the event ^
	Publicize the event and invite scientists to participate in the next benchathlon by running their CBIR algorithms and/or contributing benchmarks

Manage images and metadata ^

The old (February 2001) images are here We are currently rebuilding the image repository following the directory structure used by the University of California at Berkeley Digital Library Project; we hope this will allow us to piggy-back on their research, described here The physical location of the images is http://www.benchathlon.net/img/done/ and the next task is to install a database to capture the metadata. We can then use Berkeley's cgi script to present a brower interface to the images, which can be used for the indexing Instructions