Revision history [back]

How big is your training set? K-Means is an O(n^2) algorithm, so it can take a long time on large datasets.

For reference, I trained a similar vocabulary as you with 3138432 SIFT descriptors in it and it took around 48 hours.