Experiment 4

(1) data, codes, and readme
(2)a much faster implementation for learning
(2.1)with -log F whitening
(3)with M-step re-learning of the active basis for each cluster
(4)separating out head-shoulder examples
(5)separating horses with different directions
(6)K-mean based on active correlation
(7)separating three clusters

First cluster

Template of the first cluster

Second cluster

Template of the second cluster

Back to active basis homepage