Case 1: Horses facing two different directions (k-mean, with DoG)

Parameters Setting:
General Parameters: nOrient = 16; sizeTemplatex=60; sizeTemplatey=80; GaborScaleList=[0.8, 0.5, 0.3]; useDoG = true; isLocalNormalize = false; DoGScale = 10;
HMC Parameters: lambdaLearningRate = 0.01; epsilon = 0.01; L = 10; nIteraton =6; 12x12 chains
Clustering Parameters: #EM iteration = 20; #clusters = 2; isSoftClassification = false.

iteration 0: start from random intialization

iteration 1:

iteration 2:

iteration 3:

iteration 4:

iteration 10:

iteration 12:

The final learned templates