A search on Impressionism reveals 12 large thumbnail stacks of paintings by a variety of Impressionist painters organized by their visual and semantic similarities.  Select a stack and the images arrange themselves as a cluster, swirling into view.

You can continue to drill down through related subgroups within each cluster in a mapping strategy, reminiscent of Google's Wonder Wheel available in the Search Options panel.  Image Swirl searches are currently available for approximately 200,000 queries.

The Help page explains what goes on under the search hood:

How does this work?
Google Image Swirl combines a variety of image similarity features with additional metadata about the images to build a hierarchy of clusters of image search results.

What are the system requirements?
This demo requires Javascript to be enabled and a Flash plugin to be installed in order to work. It has been tested on Linux (Chrome and Firefox), Windows (Chrome, IE, Firefox, Safari, Opera) and Mac OS/X (Chrome, Safari, Firefox).

Why are there images in clusters which don't belong? The clustering algorithm is fully automatic, and far from perfect. In most cases, computer vision cannot tell us what actually present in an image, and without this semantic information, mistakes will happen.

Students and teachers are going to love exploring here. 

Below are a further examples of visual category-type searches that make best sense for a tool like this:

 

   
Just yesterday I blogged about Bing's new similar Visual Search.  Neither are yet fully formed, but Image Swirl appears to have far more content at this point. 

I'll keep you posted as I continue to compare.