Easy tool for evaluation of images datasets, where recall isn’t an option

Often when doing image dataset evaluation you want a quick easy way to either see how different descriptors perform or just get an evaluation measure for your system. Well I have created a tool to be able to do the former. This is a simple tool so you can select images, if that image appears in another descriptors results it is auto-magically highlighted to avoid bias or mistake. As you scroll the query image stays with you and finally you can upload a relevance result for each descriptor to the server.

image_descriptor_evaluation_tool

 

Of course to use this requires you being able to ping against my server so isn’t for public use, but is quite fun to play around with javascript and see how easy although painful it can be to create GUI’s for distributed use.

Friday, June 1, 2012 6:50:00 PM Categories: Evaluation Retrieval WebDev
Stuart James