drwnPartsInference Class Reference

Helper class for running inference in a (constellation) parts-based model. Supports linear and quadratic distance transforms for deformation costs. Computes argmax_{x,c,z} m_i(x_i, z_i) + d_i(x_i, c) + p(c) where m_i(x, 0) = matchingCost_i(x) + priorCost_i(x) m_i(x, 1) = occlusionCost + priorCost_i(x) d_i(x, c) = [dx, dy]^T fabs(x - o - c) + [dx2, dy2]^T (x - o - c).^2. More...

Public Member Functions
	drwnPartsInference (int numParts, int width, int height)
void	setCentroidPrior (const cv::Mat &centroidPrior=cv::Mat())
void	setPartCosts (int partId, const cv::Mat &matchingCost, const cv::Point &offset, const drwnDeformationCost &pairwiseCost, const cv::Mat &occlusionCost=cv::Mat())
drwnPartsAssignment	inference () const
drwnPartsAssignment	inference (const cv::Point &centroid) const
cv::Mat	energyLandscape () const

Static Public Attributes
static int	MESSAGE_PASSING_SCALE = 1
static double	DEFAULT_LAMBDA = 5.0e-3
static double	DEFAULT_OCCLUSION_COST = 0.0

Protected Member Functions
cv::Mat	computeLocationMessage (const cv::Mat &belief, const drwnDeformationCost &dcost, const cv::Mat &msgIn=cv::Mat()) const

Protected Attributes
double	_lambda
double	_occlusionCost
int	_width
int	_height
int	_nParts
cv::Mat	_centroidPrior
vector< pair< cv::Mat, cv::Mat > >	_partCosts
vector< cv::Point >	_partOffsets
vector< drwnDeformationCost >	_pairwiseCosts

Detailed Description

Helper class for running inference in a (constellation) parts-based model. Supports linear and quadratic distance transforms for deformation costs. Computes argmax_{x,c,z} m_i(x_i, z_i) + d_i(x_i, c) + p(c) where m_i(x, 0) = matchingCost_i(x) + priorCost_i(x) m_i(x, 1) = occlusionCost + priorCost_i(x) d_i(x, c) = [dx, dy]^T fabs(x - o - c) + [dx2, dy2]^T (x - o - c).^2.

---

The documentation for this class was generated from the following files:

• drwnPartsModel.h
• drwnPartsModel.cpp