Visual Vocabulary
The first step in this pipeline was to create a vocabulary of visual words using k-means as the cluster method.
The big thing here was that we used SIFT features as the underlying cluster-domain.
I chose to densely sample SIFT features across all images using a bin size of 4 and a step size of 8.
I went ahead a used all of the SIFT features available, even though it took a little while to run k-means.
All other parameters were left at their default values.