When should I use a kd-tree?

Question

The other day, I was reading about kd-trees. I was looking for a concrete and simple situation where such a data structure could be useful. Does anybody have such an example?

Answer 1

I used a kd-tree to simulate light transport through spray generated by a wave. This made it possible to render droplets of spray that weren't just illuminated by lamps, but also indirect lighting coming from other droplets.

I stored points of spray in a kd-tree and then used the kd-tree to quickly find which points were near each other, and hence which droplets of water were illuminated by each other. (Actually, it was a little more sophisticated, but the structure was still a kd-tree.) This made it possible to render a nice glow through the spray.

Here's a picture of the method in use.

Note how the spray around the lights is glowing with scattered light.

Answer 2

I've used them as an efficient way to find the nearest neighbors of a given point for machine learning.

Answer 3

They are used in raytacing to break a scene down into a couple of boxes. It speeds up collisions because if a ray doesn't collide with the bounding box, it surely doesn't collide with its contents.

Answer 4

VTK has one: http://www.vtk.org/doc/nightly/html/classvtkKdTree.html for an example.

Take a look at functions like FindClosestPoint. The class takes a comparatively long time to build given a large set of input points. But after the tree is built, functions like this one run very fast.

Answer 5

I used them in a pricing project. The object was to find the best price that matched multiple criteria.