#include <ls_minimizer.h>

Public Member Functions
	ls_minimizer (int state_size, int maximum_scalar_measure_number)
	~ls_minimizer ()
void	set_max_number_of_observations_2data_1measure (int n)
void	set_max_number_of_observations_2data_2measures (int n)
void	set_max_number_of_observations_3data_2measures (int n)
	for example a 3D point projected on a 2D plane
void	set_scales (double *scales)
void	desactivate_automated_scaling (void)
void	set_verbose_level (int vl)
void	set_user_data (int slot_index, void *ptr)
void	reset_observations (void)
	call that before adding observations for a new optimization.
bool	add_observation (function_2data_1measure f, const double d0, const double d1, const double b0)
bool	add_observation (function_2data_2measures f, const double d0, const double d1, const double b0, const double b1)
bool	add_observation (function_3data_2measures f, const double d0, const double d1, const double d2, const double b0, const double b1)
void	set_new_iteration_callback (callback_function callback)
	this callback is called each time the state vector changes.
void	set_default_c (double c)
void	set_last_observation_c (double c)
void	set_last_observation_c_max_c_min (double c_max, double c_min)
	For robust estimation using Julien' method (default = +inf for both):
void	set_last_observation_weight (double weight)
	Set observation weight (default = 1):
void	set_last_observation_confidence (double confidence)
	Set observation confidence (default = 1). Only used for PROSAC and cat tail.
int	minimize_using_steepest_descent_from (double *initial_state)
int	minimize_using_levenberg_marquardt_from (double *initial_state)
void	lm_set_max_iterations (int it)
void	lm_set_max_failures_in_a_row (int f)
	if f iterations do not decrease the function, stop.
void	lm_set_tol_cos (double t)
	stop condition. Very restrictive: 0.999 less restrictive: 0.9
bool	minimize_using_gauss_newton_from (double *initial_state)
int	minimize_using_dogleg_from (double *initial_state)
	Dogleg (not convincing).
int	minimize_using_cattail_from (double *initial_state)
void	ct_set_max_iterations (int it)
void	set_line_search_parameters (double lambda0, double k_rough, double k_fine)
double	minimize_using_prosac (prosac_function_20scalars f, int observation_type, int min_number_of_inliers, int max_number_of_iterations)
double	minimize_using_julien_method_from (double *state, int nb_steps, int nb_iterations_per_step)
void	check_jacobians_around (double *state, double state_step)
void	set_last_observation_as_outlier (void)
void	set_ground_truth (double *state)
void	compare_outliers_with_ground_truth (void)
void	compare_state_with_ground_truth (void)

Public Attributes
ls_minimizer_type *	state
	state vector

Detailed Description

Non-linear Minimizer

Author: Vincent Lepetit

Definition at line 45 of file ls_minimizer.h.

Constructor & Destructor Documentation

ls_minimizer::ls_minimizer	(	int	state_size,
		int	maximum_scalar_measure_number
	)

ls_minimizer::~ls_minimizer ( )

Member Function Documentation

bool ls_minimizer::add_observation	(	function_2data_1measure	f,
		const double	d0,
		const double	d1,
		const double	b0
	)

bool ls_minimizer::add_observation	(	function_2data_2measures	f,
		const double	d0,
		const double	d1,
		const double	b0,
		const double	b1
	)

bool ls_minimizer::add_observation	(	function_3data_2measures	f,
		const double	d0,
		const double	d1,
		const double	d2,
		const double	b0,
		const double	b1
	)

In the case of a 3d point projected on a 2d plane: d0 = X, d1 = Y, d2 = Z (b0,b1) projection goal f : f(d0,d1,d2,state, ..) is the function computing the projection and its gradient.

void ls_minimizer::check_jacobians_around	(	double *	state,
		double	state_step
	)

Checking derivatives: compares finite difference jacobian and analytical jacobian computed by the user provided functions.

void ls_minimizer::compare_outliers_with_ground_truth ( void )

void ls_minimizer::compare_state_with_ground_truth ( void )

void ls_minimizer::ct_set_max_iterations ( int it )

inline

Definition at line 140 of file ls_minimizer.h.

void ls_minimizer::desactivate_automated_scaling ( void )

inline

Definition at line 62 of file ls_minimizer.h.

void ls_minimizer::lm_set_max_failures_in_a_row ( int f )

if f iterations do not decrease the function, stop.

void ls_minimizer::lm_set_max_iterations ( int it )

void ls_minimizer::lm_set_tol_cos ( double t )

stop condition. Very restrictive: 0.999 less restrictive: 0.9

int ls_minimizer::minimize_using_cattail_from ( double * initial_state )

Cat Tail (experimental) a mix between gauss-newton and line search.

int ls_minimizer::minimize_using_dogleg_from ( double * initial_state )

Dogleg (not convincing).

bool ls_minimizer::minimize_using_gauss_newton_from ( double * initial_state )

double ls_minimizer::minimize_using_julien_method_from	(	double *	state,
		int	nb_steps,
		int	nb_iterations_per_step
	)

int ls_minimizer::minimize_using_levenberg_marquardt_from ( double * initial_state )

double ls_minimizer::minimize_using_prosac	(	prosac_function_20scalars	f,
		int	observation_type,
		int	min_number_of_inliers,
		int	max_number_of_iterations
	)

int ls_minimizer::minimize_using_steepest_descent_from ( double * initial_state )

void ls_minimizer::reset_observations ( void )

call that before adding observations for a new optimization.

void ls_minimizer::set_default_c ( double c )

For robust estimation (default = +inf) residual = max( ||b-f(d,state)||^2, c^2 ) This value will be applied to next added observations

void ls_minimizer::set_ground_truth ( double * state )

void ls_minimizer::set_last_observation_as_outlier ( void )

void ls_minimizer::set_last_observation_c ( double c )

For robust estimation (default = +inf): change the last added observation robust estimator value

void ls_minimizer::set_last_observation_c_max_c_min	(	double	c_max,
		double	c_min
	)

For robust estimation using Julien' method (default = +inf for both):

void ls_minimizer::set_last_observation_confidence ( double confidence )

Set observation confidence (default = 1). Only used for PROSAC and cat tail.

void ls_minimizer::set_last_observation_weight ( double weight )

Set observation weight (default = 1):

void ls_minimizer::set_line_search_parameters	(	double	lambda0,
		double	k_rough,
		double	k_fine
	)

void ls_minimizer::set_max_number_of_observations_2data_1measure ( int n )

void ls_minimizer::set_max_number_of_observations_2data_2measures ( int n )

void ls_minimizer::set_max_number_of_observations_3data_2measures ( int n )

for example a 3D point projected on a 2D plane

void ls_minimizer::set_new_iteration_callback ( callback_function callback )

this callback is called each time the state vector changes.

void ls_minimizer::set_scales ( double * scales )

Scale applied to each optimized parameter (each dimension of the optimization space). Not used by default, but an automated method is used.

void ls_minimizer::set_user_data	(	int	slot_index,
		void *	ptr
	)

user arbitrary pointers. slot_index should be in [0,9] this pointer array is passed to the callback functions

void ls_minimizer::set_verbose_level ( int vl )

0 = nothing 1 = messages before and after minimization 2 = message at each iteration. 3 = debug

Member Data Documentation

ls_minimizer_type* ls_minimizer::state

state vector

Definition at line 116 of file ls_minimizer.h.

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

/home/adrien/Téléchargements/bazar/bazar-1.3.1/starter/optimization/ls_minimizer.h

Public Member Functions

Public Attributes

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation