GitHub - zzsi/EMClusterAligned: clustering for social role data

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 6 Commits
positiveImage		positiveImage
ApplyFilterfftSame.m		ApplyFilterfftSame.m
CorrFilter.m		CorrFilter.m
EMClusterPerRect.m		EMClusterPerRect.m
ExponentialModel.m		ExponentialModel.m
GaborFilter.m		GaborFilter.m
MakeFilter.m		MakeFilter.m
ParameterCodeImage.m		ParameterCodeImage.m
README		README
SUM1MAX1.m		SUM1MAX1.m
StartDP.m		StartDP.m
StartFromHere.m		StartFromHere.m
displayImages.m		displayImages.m
displayMatchedTemplate.m		displayMatchedTemplate.m
displayTemplate.m		displayTemplate.m
drawGaborSymbol.m		drawGaborSymbol.m
findRect.m		findRect.m
inscribe.m		inscribe.m
mexc_CollectSUM1Maps.cpp		mexc_CollectSUM1Maps.cpp
mexc_ComputeMAX1.cpp		mexc_ComputeMAX1.cpp
mexc_ComputeSUMMAX2.cpp		mexc_ComputeSUMMAX2.cpp
mexc_ComputeSUMMAX2FixCluster.cpp		mexc_ComputeSUMMAX2FixCluster.cpp
mexc_CropInstanceNew.cpp		mexc_CropInstanceNew.cpp
mexc_Histogram.cpp		mexc_Histogram.cpp
mexc_LocalNormalize.cpp		mexc_LocalNormalize.cpp
mexc_SharedSketch.cpp		mexc_SharedSketch.cpp
mexc_Sigmoid.cpp		mexc_Sigmoid.cpp
mexc_TemplateAffineTransform.cpp		mexc_TemplateAffineTransform.cpp
storedExponentialModel1.mat		storedExponentialModel1.mat

Repository files navigation

Functionality:

	Learn a re-configurable AND-OR template for input images.

	It performs two steps:
	1) EM clustering for rectangular regions inside the image lattice. The images are assumed to be aligned (not necessarily the same size).
	2) Learning an optimal partition of the whole object into rectangular parts (partition configuration) using dynamic programming.

Run StartFromHere.m in the Matlab. This produces the visual codebook for rectangular patches. And you can monitor intermediate results in output/.
Then run StartDP.m to learn the partition configuration.