Detailed Description

template<class CAMERA>
class gtsam::SmartFactorBase< CAMERA >

Base class for smart factors.

This base class has no internal point, but it has a measurement, noise model and an optional sensor pose. This class mainly computes the derivatives and returns them as a variety of factors. The methods take a CameraSet<CAMERA> argument and the value of a point, which is kept in the derived class.

Template Parameters

CAMERA should behave like a PinholeCamera.

Inheritance diagram for gtsam::SmartFactorBase< CAMERA >:

Public Member Functions
	SmartFactorBase ()
	Default Constructor, for serialization.

	SmartFactorBase (const SharedNoiseModel &sharedNoiseModel, boost::optional< Pose3 > body_P_sensor=boost::none, size_t expectedNumberCameras=10)
	Construct with given noise model and optional arguments.

	~SmartFactorBase () override
	Virtual destructor, subclasses from NonlinearFactor.

void	add (const Z &measured, const Key &key)
	Add a new measurement and pose/camera key. More...

void	add (const ZVector &measurements, const KeyVector &cameraKeys)
	Add a bunch of measurements, together with the camera keys.

template<class SFM_TRACK >
void	add (const SFM_TRACK &trackToAdd)
	Add an entire SfM_track (collection of cameras observing a single point). More...

size_t	dim () const override
	Return the dimension (number of rows!) of the factor. More...

const ZVector &	measured () const
	Return the 2D measurements (ZDim, in general).

virtual Cameras	cameras (const Values &values) const
	Collect all cameras: important that in key order. More...

void	print (const std::string &s="", const KeyFormatter &keyFormatter=DefaultKeyFormatter) const override
	print More...

bool	equals (const NonlinearFactor &p, double tol=1e-9) const override
	equals More...

template<class POINT >
Vector	unwhitenedError (const Cameras &cameras, const POINT &point, boost::optional< typename Cameras::FBlocks & > Fs=boost::none, boost::optional< Matrix & > E=boost::none) const
	Compute reprojection errors [h(x)-z] = [cameras.project(p)-z] and derivatives. More...

virtual void	correctForMissingMeasurements (const Cameras &cameras, Vector &ue, boost::optional< typename Cameras::FBlocks & > Fs=boost::none, boost::optional< Matrix & > E=boost::none) const
	This corrects the Jacobians for the case in which some 2D measurement is missing (nan). More...

template<class POINT >
Vector	whitenedError (const Cameras &cameras, const POINT &point) const
	Calculate vector of re-projection errors [h(x)-z] = [cameras.project(p) - z], with the noise model applied.

template<class POINT >
double	totalReprojectionError (const Cameras &cameras, const POINT &point) const
	Calculate the error of the factor. More...

template<class POINT >
void	computeJacobians (FBlocks &Fs, Matrix &E, Vector &b, const Cameras &cameras, const POINT &point) const
	Compute F, E, and b (called below in both vanilla and SVD versions), where F is a vector of derivatives wrpt the cameras, and E the stacked derivatives with respect to the point. More...

template<class POINT >
void	computeJacobiansSVD (FBlocks &Fs, Matrix &Enull, Vector &b, const Cameras &cameras, const POINT &point) const
	SVD version that produces smaller Jacobian matrices by doing an SVD decomposition on E, and returning the left nulkl-space of E. More...

boost::shared_ptr< RegularHessianFactor< Dim > >	createHessianFactor (const Cameras &cameras, const Point3 &point, const double lambda=0.0, bool diagonalDamping=false) const
	Linearize to a Hessianfactor.

void	updateAugmentedHessian (const Cameras &cameras, const Point3 &point, const double lambda, bool diagonalDamping, SymmetricBlockMatrix &augmentedHessian, const KeyVector allKeys) const
	Add the contribution of the smart factor to a pre-allocated Hessian, using sparse linear algebra. More...

void	whitenJacobians (FBlocks &F, Matrix &E, Vector &b) const
	Whiten the Jacobians computed by computeJacobians using noiseModel_.

boost::shared_ptr< RegularImplicitSchurFactor< CAMERA > >	createRegularImplicitSchurFactor (const Cameras &cameras, const Point3 &point, double lambda=0.0, bool diagonalDamping=false) const
	Return Jacobians as RegularImplicitSchurFactor with raw access.

boost::shared_ptr< JacobianFactorQ< Dim, ZDim > >	createJacobianQFactor (const Cameras &cameras, const Point3 &point, double lambda=0.0, bool diagonalDamping=false) const
	Return Jacobians as JacobianFactorQ.

boost::shared_ptr< JacobianFactor >	createJacobianSVDFactor (const Cameras &cameras, const Point3 &point, double lambda=0.0) const
	Return Jacobians as JacobianFactorSVD. More...

Pose3	body_P_sensor () const

Public Member Functions inherited from gtsam::NonlinearFactor
	NonlinearFactor ()
	Default constructor for I/O only.

template<typename CONTAINER >
	NonlinearFactor (const CONTAINER &keys)
	Constructor from a collection of the keys involved in this factor.

virtual	~NonlinearFactor ()
	Destructor.

virtual double	error (const Values &c) const =0
	Calculate the error of the factor This is typically equal to log-likelihood, e.g. More...

virtual bool	active (const Values &) const
	Checks whether a factor should be used based on a set of values. More...

virtual boost::shared_ptr< GaussianFactor >	linearize (const Values &c) const =0
	linearize to a GaussianFactor More...

virtual shared_ptr	clone () const
	Creates a shared_ptr clone of the factor - needs to be specialized to allow for subclasses. More...

virtual shared_ptr	rekey (const std::map< Key, Key > &rekey_mapping) const
	Creates a shared_ptr clone of the factor with different keys using a map from old->new keys. More...

virtual shared_ptr	rekey (const KeyVector &new_keys) const
	Clones a factor and fully replaces its keys. More...

virtual bool	sendable () const
	Should the factor be evaluated in the same thread as the caller This is to enable factors that has shared states (like the Python GIL lock) More...

Public Member Functions inherited from gtsam::Factor
virtual	~Factor ()=default
	Default destructor.

KeyVector &	keys ()

iterator	begin ()
	Iterator at beginning of involved variable keys.

iterator	end ()
	Iterator at end of involved variable keys.

virtual void	printKeys (const std::string &s="Factor", const KeyFormatter &formatter=DefaultKeyFormatter) const
	print only keys More...

Key	front () const
	First key.

Key	back () const
	Last key.

const_iterator	find (Key key) const
	find

const KeyVector &	keys () const
	Access the factor's involved variable keys.

const_iterator	begin () const
	Iterator at beginning of involved variable keys.

const_iterator	end () const
	Iterator at end of involved variable keys.

size_t	size () const

Static Public Member Functions
static Matrix	PointCov (const Matrix &E)
	Computes Point Covariance P from the "point Jacobian" E.

static void	FillDiagonalF (const FBlocks &Fs, Matrix &F)
	Create BIG block-diagonal matrix F from Fblocks.

Public Attributes
GTSAM_MAKE_ALIGNED_OPERATOR_NEW typedef boost::shared_ptr< This >	shared_ptr
	shorthand for a smart pointer to a factor.

Static Public Attributes
static const int	Dim = traits<CAMERA>::dimension
	Camera dimension.

static const int	ZDim = traits<Z>::dimension
	Measurement dimension.

Public Types
typedef Eigen::Matrix< double, ZDim, Dim >	MatrixZD

typedef std::vector< MatrixZD, Eigen::aligned_allocator< MatrixZD > >	FBlocks

typedef CameraSet< CAMERA >	Cameras
	The CameraSet data structure is used to refer to a set of cameras.

Public Types inherited from gtsam::NonlinearFactor
typedef boost::shared_ptr< This >	shared_ptr

Public Types inherited from gtsam::Factor
typedef KeyVector::iterator	iterator
	Iterator over keys.

typedef KeyVector::const_iterator	const_iterator
	Const iterator over keys.

Protected Attributes
SharedIsotropic	noiseModel_
	As of Feb 22, 2015, the noise model is the same for all measurements and is isotropic. More...

ZVector	measured_
	Measurements for each of the m views. More...

boost::optional< Pose3 >	body_P_sensor_
	Pose of the camera in the body frame.

FBlocks	Fs

Protected Attributes inherited from gtsam::Factor
KeyVector	keys_
	The keys involved in this factor.

Friends
class	boost::serialization::access
	Serialization function.

Additional Inherited Members
Protected Types inherited from gtsam::NonlinearFactor
typedef Factor	Base

typedef NonlinearFactor	This

Protected Member Functions inherited from gtsam::Factor
	Factor ()
	Default constructor for I/O.

template<typename CONTAINER >
	Factor (const CONTAINER &keys)
	Construct factor from container of keys. More...

template<typename ITERATOR >
	Factor (ITERATOR first, ITERATOR last)
	Construct factor from iterator keys. More...

bool	equals (const This &other, double tol=1e-9) const
	check equality

Static Protected Member Functions inherited from gtsam::Factor
template<typename CONTAINER >
static Factor	FromKeys (const CONTAINER &keys)
	Construct factor from container of keys. More...

template<typename ITERATOR >
static Factor	FromIterators (ITERATOR first, ITERATOR last)
	Construct factor from iterator keys. More...

Member Function Documentation

◆ add() [1/2]

template<class CAMERA >

template<class SFM_TRACK >

void gtsam::SmartFactorBase< CAMERA >::add ( const SFM_TRACK & trackToAdd )

inline

Add an entire SfM_track (collection of cameras observing a single point).

The noise is assumed to be the same for all measurements.

◆ add() [2/2]

template<class CAMERA >

void gtsam::SmartFactorBase< CAMERA >::add	(	const Z &	measured,
		const Key &	key
	)

inline

Add a new measurement and pose/camera key.

Parameters

measured	is the 2m dimensional projection of a single landmark
key	is the index corresponding to the camera observing the landmark

◆ cameras()

template<class CAMERA >

virtual Cameras gtsam::SmartFactorBase< CAMERA >::cameras ( const Values & values ) const

inlinevirtual

Collect all cameras: important that in key order.

Reimplemented in gtsam::SmartProjectionPoseFactor< CALIBRATION >, gtsam::SmartProjectionRigFactor< CAMERA >, gtsam::SmartProjectionPoseFactorRollingShutter< CAMERA >, gtsam::SmartStereoProjectionFactorPP, and gtsam::SmartStereoProjectionPoseFactor.

◆ computeJacobians()

template<class CAMERA >

template<class POINT >

void gtsam::SmartFactorBase< CAMERA >::computeJacobians	(	FBlocks &	Fs,
		Matrix &	E,
		Vector &	b,
		const Cameras &	cameras,
		const POINT &	point
	)		const

inline

Compute F, E, and b (called below in both vanilla and SVD versions), where F is a vector of derivatives wrpt the cameras, and E the stacked derivatives with respect to the point.

The value of cameras/point are passed as parameters.

◆ computeJacobiansSVD()

template<class CAMERA >

template<class POINT >

void gtsam::SmartFactorBase< CAMERA >::computeJacobiansSVD	(	FBlocks &	Fs,
		Matrix &	Enull,
		Vector &	b,
		const Cameras &	cameras,
		const POINT &	point
	)		const

inline

SVD version that produces smaller Jacobian matrices by doing an SVD decomposition on E, and returning the left nulkl-space of E.

See JacobianFactorSVD for more documentation.

◆ correctForMissingMeasurements()

template<class CAMERA >

virtual void gtsam::SmartFactorBase< CAMERA >::correctForMissingMeasurements	(	const Cameras &	cameras,
		Vector &	ue,
		boost::optional< typename Cameras::FBlocks & >	Fs = `boost::none`,
		boost::optional< Matrix & >	E = `boost::none`
	)		const

inlinevirtual

This corrects the Jacobians for the case in which some 2D measurement is missing (nan).

In practice, this does not do anything in the monocular case, but it is implemented in the stereo version.

Reimplemented in gtsam::SmartStereoProjectionFactor.

◆ createJacobianSVDFactor()

template<class CAMERA >

boost::shared_ptr< JacobianFactor > gtsam::SmartFactorBase< CAMERA >::createJacobianSVDFactor	(	const Cameras &	cameras,
		const Point3 &	point,
		double	lambda = `0.0`
	)		const

inline

Return Jacobians as JacobianFactorSVD.

TODO(dellaert): lambda is currently ignored

◆ dim()

template<class CAMERA >

size_t gtsam::SmartFactorBase< CAMERA >::dim ( ) const

inlineoverridevirtual

Return the dimension (number of rows!) of the factor.

Implements gtsam::NonlinearFactor.

◆ equals()

template<class CAMERA >

bool gtsam::SmartFactorBase< CAMERA >::equals	(	const NonlinearFactor &	p,
		double	tol = `1e-9`
	)		const

inlineoverridevirtual

equals

Reimplemented from gtsam::NonlinearFactor.

Reimplemented in gtsam::SmartProjectionFactor< CAMERA >, gtsam::SmartProjectionFactor< PinholePose< CALIBRATION > >, gtsam::SmartProjectionRigFactor< CAMERA >, and gtsam::SmartProjectionPoseFactorRollingShutter< CAMERA >.

◆ print()

template<class CAMERA >

void gtsam::SmartFactorBase< CAMERA >::print	(	const std::string &	s = `""`,
		const KeyFormatter &	keyFormatter = `DefaultKeyFormatter`
	)		const

inlineoverridevirtual

print

Parameters

s	optional string naming the factor
keyFormatter	optional formatter useful for printing Symbols

Reimplemented from gtsam::NonlinearFactor.

Reimplemented in gtsam::SmartProjectionFactor< CAMERA >, gtsam::SmartProjectionFactor< PinholePose< CALIBRATION > >, gtsam::SmartProjectionRigFactor< CAMERA >, and gtsam::SmartProjectionPoseFactorRollingShutter< CAMERA >.

◆ totalReprojectionError()

template<class CAMERA >

template<class POINT >

double gtsam::SmartFactorBase< CAMERA >::totalReprojectionError	(	const Cameras &	cameras,
		const POINT &	point
	)		const

inline

Calculate the error of the factor.

This is the log-likelihood, e.g. \( 0.5(h(x)-z)^2/\sigma^2 \) in case of Gaussian. In this class, we take the raw prediction error \( h(x)-z \), ask the noise model to transform it to \( (h(x)-z)^2/\sigma^2 \), and then multiply by 0.5. Will be used in "error(Values)" function required by NonlinearFactor base class

◆ unwhitenedError()

template<class CAMERA >

template<class POINT >

Vector gtsam::SmartFactorBase< CAMERA >::unwhitenedError	(	const Cameras &	cameras,
		const POINT &	point,
		boost::optional< typename Cameras::FBlocks & >	Fs = `boost::none`,
		boost::optional< Matrix & >	E = `boost::none`
	)		const

inline

Compute reprojection errors [h(x)-z] = [cameras.project(p)-z] and derivatives.

This is the error before the noise model is applied.

◆ updateAugmentedHessian()

template<class CAMERA >

void gtsam::SmartFactorBase< CAMERA >::updateAugmentedHessian	(	const Cameras &	cameras,
		const Point3 &	point,
		const double	lambda,
		bool	diagonalDamping,
		SymmetricBlockMatrix &	augmentedHessian,
		const KeyVector	allKeys
	)		const

inline

Add the contribution of the smart factor to a pre-allocated Hessian, using sparse linear algebra.

More efficient than the creation of the Hessian without preallocation of the SymmetricBlockMatrix

Member Data Documentation

◆ measured_

template<class CAMERA >

ZVector gtsam::SmartFactorBase< CAMERA >::measured_

protected

Measurements for each of the m views.

We keep a copy of the measurements for I/O and computing the error. The order is kept the same as the keys that we use to create the factor.

◆ noiseModel_

template<class CAMERA >

SharedIsotropic gtsam::SmartFactorBase< CAMERA >::noiseModel_

protected

As of Feb 22, 2015, the noise model is the same for all measurements and is isotropic.

This allows for moving most calculations of Schur complement etc. to be easily moved to CameraSet, and also agrees pragmatically with what is normally done.

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

/Users/dellaert/git/github/gtsam/slam/SmartFactorBase.h

Detailed Description

Public Member Functions

Static Public Member Functions

Public Attributes

Static Public Attributes

Public Types

Protected Attributes

Friends

Additional Inherited Members

Member Function Documentation

◆ add() [1/2]

◆ add() [2/2]

◆ cameras()

◆ computeJacobians()

◆ computeJacobiansSVD()

◆ correctForMissingMeasurements()

◆ createJacobianSVDFactor()

◆ dim()

◆ equals()

◆ print()

◆ totalReprojectionError()

◆ unwhitenedError()

◆ updateAugmentedHessian()

Member Data Documentation

◆ measured_

◆ noiseModel_