#include <psi_likelihood.h>

Public Member Functions
	Frame_likelihood (size_t width, size_t height, const boost::function2< double, const Bbox &, const Bbox & > &pairwise_box_likelihood, const boost::function1< double, const Bbox & > &noise_box_likelihood, double p_noise)

double	operator() (const std::vector< Bbox > &data_boxes, const std::vector< Bbox > &model_boxes) const

double	compute_visibility (const std::vector< Bbox > &all_boxes, size_t box_i) const

double	p_missing (double visibility) const

double	p_correspondence (size_t num_data, size_t num_noise) const

std::vector< double >	get_p_missing (const std::vector< Bbox > &model) const

kjb::Matrix	get_affinity (const std::vector< Bbox > &data, const std::vector< Bbox > &model) const

std::vector< size_t >	get_correspondence_greedily (const kjb::Matrix &affinities) const

Frame_likelihood &	dont_marginalize_noise ()

Protected Member Functions
double	operator() (const std::vector< Bbox > &data_boxes, const std::vector< Bbox > &models, const kjb::Matrix &affinities, const std::vector< size_t > &correspondences, const std::vector< double > &m_p_missing) const

kjb::Matrix	get_affinity (const std::vector< Bbox > &data, const std::vector< Bbox > &model, const std::vector< double > &m_p_missing) const

Constructor & Destructor Documentation

kjb::psi::Frame_likelihood::Frame_likelihood	(	size_t	width,
		size_t	height,
		const boost::function2< double, const Bbox &, const Bbox & > &	pairwise_box_likelihood,
		const boost::function1< double, const Bbox & > &	noise_box_likelihood,
		double	p_noise
	)

inline

Parameters

p_noise bernoulli noise process

Member Function Documentation

double kjb::psi::Frame_likelihood::compute_visibility	(	const std::vector< Bbox > &	all_boxes,
		size_t	box_i
	)		const

Find the amount of a bounding box that is visible, from zero to 1. Reduction in visibility can be the result of occlusion from other boxes in front of it, or from the box being parially (or completely) outside the view of the camera.

Currently, this occlusion approximated by discretizing the box into 64 cells, and doing a binary test on each cell to see if it's occluded. As a result, the returned values will have a resolution of 1/64, but it gives a running time of O(n) instead of O(n^2 log n) for an exact solution (and less error-prone!!).

Parameters

all_boxes	The set of all bounding boxes, sorted in order of depth, with nearest first.
i	the index of the box to find the occlusion of

Precondition: all_boxes is sorted descending by depth

Frame_likelihood& kjb::psi::Frame_likelihood::dont_marginalize_noise ( )

inline

You'll almost never call this. This sets a little flag that lets the likelihood demo be more illustrative by allowing the correspondence matching to switch from "match" to "noise" if the match is too bad.

kjb::Matrix kjb::psi::Frame_likelihood::get_affinity	(	const std::vector< Bbox > &	data,
		const std::vector< Bbox > &	model
	)		const

inline

kjb::Matrix kjb::psi::Frame_likelihood::get_affinity	(	const std::vector< Bbox > &	data,
		const std::vector< Bbox > &	model,
		const std::vector< double > &	m_p_missing
	)		const

protected

Precondition: model is sorted descending by depth

Parameters

data	Data boudning boxes
model	Bounding boxes generated by the model
m_p_missing	Probability of each model being undetected (based on occlusion/visibility)

std::vector< size_t > kjb::psi::Frame_likelihood::get_correspondence_greedily ( const kjb::Matrix & affinities ) const

std::vector< double > kjb::psi::Frame_likelihood::get_p_missing ( const std::vector< Bbox > & model ) const

Returns a vector containing the probability of each model being undetected, based on how much it is occluded by the others and whether or not it is on the screen

double kjb::psi::Frame_likelihood::operator()	(	const std::vector< Bbox > &	data_boxes,
		const std::vector< Bbox > &	model_boxes
	)		const

inline

Compare two sets of bounding boxes.

This will find a correspodence between data and model by greedilly matching data boxes to model boxes or noise.

Precondition: model_boxes are sorted in order of depth, with nearest first

double kjb::psi::Frame_likelihood::operator()	(	const std::vector< Bbox > &	data_boxes,
		const std::vector< Bbox > &	model_boxes,
		const kjb::Matrix &	affinities,
		const std::vector< size_t > &	correspondences,
		const std::vector< double > &	m_p_missing
	)		const

protected

Parameters

affinities	The affinity matrix returned by get_affinities().
correspondences	The correpondence map returned by get_correspondences_greedily()

Precondition: model_boxes are sorted descending by depth

double kjb::psi::Frame_likelihood::p_correspondence	(	size_t	num_data,
		size_t	num_noise
	)		const

Parameters

num_data	the total number of data boxes
num_noise	the number of data boxes assigned to noise

Precondition: the correspondence is already "valid", i.e. one-to-one plus a noies bucket

double kjb::psi::Frame_likelihood::p_missing ( double visibility ) const

inline

Parameters

visibility the amount of a box's area that is visible by the camera

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

/work/kobus/src/lib/psi_cpp/psi_likelihood.h
/work/kobus/src/lib/psi_cpp/psi_likelihood.cpp

Public Member Functions

Protected Member Functions

Constructor & Destructor Documentation

Member Function Documentation