This is a simple and flexible implementation of a randomized decision forest that supports configurable
decision functions, forest size, pruning types, and tree depth. Training and classification data is provided
as simple 2D images with support for up to 256 classification labels.
I became interested in this technology while working on Kinect for Xbox 360. Kinect leverages randomized decision forests as a key component of its skeletal tracking pipeline. This algorithm is highly parallelizable and capable of solving many different kinds of pattern recognition problems.
I'm currently working on open sourcing this project. Stay tuned!
For more information check out my blog for posts related to machine learning and image processing.