SLDMRF

Spatial Latent Dirichlet Markov Random Fields

% --------------------------------------------------------------

% author: Hanchen Xiong @ University of Innsbruck

% --------------------------------------------------------------

Please cite the paper if you use the code in your work: Hanchen Xiong, Sandor Szedmak, Justus Piater, 3D Object Class Geometry Modeling with Spatial Latent Dirichlet Markov Random Fields. 35th German Conference on Pattern Recognition, pp. 51–60, 2013. Springer LNCS 8142.

Application of SLDMRF for 3D part-based object modeling.

% a simple example to use the code 

% do alignment for all instances within the same category 
[aligned_point_files, pre_aligned_point_files] = category_alignment('/Category_point/motors_m');
% constuct 3D vorcabulary and discrete MRF, by default a 20*20*20 size bounding box is used
[documents,connections,min_3D,interval] = discretization(aligned_point_files);
% the number of segments 
K = 5; 
% number of 3D visual words
V = 20^3; 
[labels,psi,theta]=sldmrf_gibbs_sampler(documents,connections,K,V);

% visualize the segmentations
display_labels_3D(labels,aligned_point_files);

% visualize probabilistic part models
display_theta_3D(theta);