gtsam/gtsam/linear/LossFunctions.h

458 lines
16 KiB
C++

/* ----------------------------------------------------------------------------
* GTSAM Copyright 2010, Georgia Tech Research Corporation,
* Atlanta, Georgia 30332-0415
* All Rights Reserved
* Authors: Frank Dellaert, et al. (see THANKS for the full author list)
* See LICENSE for the license information
* -------------------------------------------------------------------------- */
/**
* @file NoiseModel.h
* @date Jan 13, 2010
* @author Richard Roberts
* @author Frank Dellaert
*/
#pragma once
#include <gtsam/base/Matrix.h>
#include <gtsam/base/Testable.h>
#include <gtsam/dllexport.h>
#include <boost/serialization/extended_type_info.hpp>
#include <boost/serialization/nvp.hpp>
#include <boost/serialization/version.hpp>
#include <boost/serialization/optional.hpp>
#include <boost/serialization/shared_ptr.hpp>
#include <boost/serialization/singleton.hpp>
namespace gtsam {
namespace noiseModel {
// clang-format off
/**
* The mEstimator name space contains all robust error functions.
* It mirrors the exposition at
* https://members.loria.fr/MOBerger/Enseignement/Master2/Documents/ZhangIVC-97-01.pdf
* which talks about minimizing \sum \rho(r_i), where \rho is a loss function of choice.
*
* To illustrate, let's consider the least-squares (L2), L1, and Huber estimators as examples:
*
* Name Symbol Least-Squares L1-norm Huber
* Loss \rho(x) 0.5*x^2 |x| 0.5*x^2 if |x|<k, 0.5*k^2 + k|x-k| otherwise
* Derivative \phi(x) x sgn(x) x if |x|<k, k sgn(x) otherwise
* Weight w(x)=\phi(x)/x 1 1/|x| 1 if |x|<k, k/|x| otherwise
*
* With these definitions, D(\rho(x), p) = \phi(x) D(x,p) = w(x) x D(x,p) = w(x) D(L2(x), p),
* and hence we can solve the equivalent weighted least squares problem \sum w(r_i) \rho(r_i)
*
* Each M-estimator in the mEstimator name space simply implements the above functions.
*/
// clang-format on
namespace mEstimator {
/**
* Pure virtual class for all robust error function classes.
*
* It provides the machinery for block vs scalar reweighting strategies, in
* addition to defining the interface of derived classes.
*/
class GTSAM_EXPORT Base {
public:
/** the rows can be weighted independently according to the error
* or uniformly with the norm of the right hand side */
enum ReweightScheme { Scalar, Block };
typedef std::shared_ptr<Base> shared_ptr;
protected:
/// Strategy for reweighting \sa ReweightScheme
ReweightScheme reweight_;
public:
Base(const ReweightScheme reweight = Block) : reweight_(reweight) {}
virtual ~Base() {}
/// Returns the reweight scheme, as explained in ReweightScheme
ReweightScheme reweightScheme() const { return reweight_; }
/**
* This method is responsible for returning the total penalty for a given
* amount of error. For example, this method is responsible for implementing
* the quadratic function for an L2 penalty, the absolute value function for
* an L1 penalty, etc.
*
* TODO(mikebosse): When the loss function has as input the norm of the
* error vector, then it prevents implementations of asymmeric loss
* functions. It would be better for this function to accept the vector and
* internally call the norm if necessary.
*
* This returns \rho(x) in \ref mEstimator
*/
virtual double loss(double distance) const { return 0; }
/**
* This method is responsible for returning the weight function for a given
* amount of error. The weight function is related to the analytic derivative
* of the loss function. See
* https://members.loria.fr/MOBerger/Enseignement/Master2/Documents/ZhangIVC-97-01.pdf
* for details. This method is required when optimizing cost functions with
* robust penalties using iteratively re-weighted least squares.
*
* This returns w(x) in \ref mEstimator
*/
virtual double weight(double distance) const = 0;
virtual void print(const std::string &s) const = 0;
virtual bool equals(const Base &expected, double tol = 1e-8) const = 0;
double sqrtWeight(double distance) const { return std::sqrt(weight(distance)); }
/** produce a weight vector according to an error vector and the implemented
* robust function */
Vector weight(const Vector &error) const;
/** square root version of the weight function */
Vector sqrtWeight(const Vector &error) const {
return weight(error).cwiseSqrt();
}
/** reweight block matrices and a vector according to their weight
* implementation */
void reweight(Vector &error) const;
void reweight(std::vector<Matrix> &A, Vector &error) const;
void reweight(Matrix &A, Vector &error) const;
void reweight(Matrix &A1, Matrix &A2, Vector &error) const;
void reweight(Matrix &A1, Matrix &A2, Matrix &A3, Vector &error) const;
private:
/** Serialization function */
#ifdef GTSAM_ENABLE_BOOST_SERIALIZATION
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_NVP(reweight_);
}
#endif
};
/** "Null" robust loss function, equivalent to a Gaussian pdf noise model, or
* plain least-squares (non-robust).
*
* This model has no additional parameters.
*
* - Loss \rho(x) = 0.5 x²
* - Derivative \phi(x) = x
* - Weight w(x) = \phi(x)/x = 1
*/
class GTSAM_EXPORT Null : public Base {
public:
typedef std::shared_ptr<Null> shared_ptr;
Null(const ReweightScheme reweight = Block) : Base(reweight) {}
~Null() override {}
double weight(double /*error*/) const override { return 1.0; }
double loss(double distance) const override { return 0.5 * distance * distance; }
void print(const std::string &s) const override;
bool equals(const Base & /*expected*/, double /*tol*/) const override { return true; }
static shared_ptr Create();
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
}
};
/** Implementation of the "Fair" robust error model (Zhang97ivc)
*
* This model has a scalar parameter "c".
*
* - Loss \rho(x) = c² (|x|/c - log(1+|x|/c))
* - Derivative \phi(x) = x/(1+|x|/c)
* - Weight w(x) = \phi(x)/x = 1/(1+|x|/c)
*/
class GTSAM_EXPORT Fair : public Base {
protected:
double c_;
public:
typedef std::shared_ptr<Fair> shared_ptr;
Fair(double c = 1.3998, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double c, const ReweightScheme reweight = Block);
double modelParameter() const { return c_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(c_);
}
};
/** The "Huber" robust error model (Zhang97ivc).
*
* This model has a scalar parameter "k".
*
* - Loss \rho(x) = 0.5 x² if |x|<k, 0.5 k² + k|x-k| otherwise
* - Derivative \phi(x) = x if |x|<k, k sgn(x) otherwise
* - Weight w(x) = \phi(x)/x = 1 if |x|<k, k/|x| otherwise
*/
class GTSAM_EXPORT Huber : public Base {
protected:
double k_;
public:
typedef std::shared_ptr<Huber> shared_ptr;
Huber(double k = 1.345, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return k_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(k_);
}
};
/** Implementation of the "Cauchy" robust error model (Lee2013IROS).
* Contributed by:
* Dipl.-Inform. Jan Oberlaender (M.Sc.), FZI Research Center for
* Information Technology, Karlsruhe, Germany.
* oberlaender@fzi.de
* Thanks Jan!
*
* This model has a scalar parameter "k".
*
* - Loss \rho(x) = 0.5 k² log(1+x²/k²)
* - Derivative \phi(x) = (k²x)/(x²+k²)
* - Weight w(x) = \phi(x)/x = k²/(x²+k²)
*/
class GTSAM_EXPORT Cauchy : public Base {
protected:
double k_, ksquared_;
public:
typedef std::shared_ptr<Cauchy> shared_ptr;
Cauchy(double k = 0.1, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return k_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(k_);
ar &BOOST_SERIALIZATION_NVP(ksquared_);
}
};
/** Implementation of the "Tukey" robust error model (Zhang97ivc).
*
* This model has a scalar parameter "c".
*
* - Loss \rho(x) = c² (1 - (1-x²/c²)³)/6 if |x|<c, c²/6 otherwise
* - Derivative \phi(x) = x(1-x²/c²)² if |x|<c, 0 otherwise
* - Weight w(x) = \phi(x)/x = (1-x²/c²)² if |x|<c, 0 otherwise
*/
class GTSAM_EXPORT Tukey : public Base {
protected:
double c_, csquared_;
public:
typedef std::shared_ptr<Tukey> shared_ptr;
Tukey(double c = 4.6851, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return c_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(c_);
}
};
/** Implementation of the "Welsch" robust error model (Zhang97ivc).
*
* This model has a scalar parameter "c".
*
* - Loss \rho(x) = -0.5 c² (exp(-x²/c²) - 1)
* - Derivative \phi(x) = x exp(-x²/c²)
* - Weight w(x) = \phi(x)/x = exp(-x²/c²)
*/
class GTSAM_EXPORT Welsch : public Base {
protected:
double c_, csquared_;
public:
typedef std::shared_ptr<Welsch> shared_ptr;
Welsch(double c = 2.9846, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return c_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(c_);
ar &BOOST_SERIALIZATION_NVP(csquared_);
}
};
/** Implementation of the "Geman-McClure" robust error model (Zhang97ivc).
*
* Note that Geman-McClure weight function uses the parameter c == 1.0,
* but here it's allowed to use different values, so we actually have
* the generalized Geman-McClure from (Agarwal15phd).
*
* - Loss \rho(x) = 0.5 (c²x²)/(c²+x²)
* - Derivative \phi(x) = xc⁴/(c²+x²)²
* - Weight w(x) = \phi(x)/x = c⁴/(c²+x²)²
*/
class GTSAM_EXPORT GemanMcClure : public Base {
public:
typedef std::shared_ptr<GemanMcClure> shared_ptr;
GemanMcClure(double c = 1.0, const ReweightScheme reweight = Block);
~GemanMcClure() override {}
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return c_; }
protected:
double c_;
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(c_);
}
};
/** DCS implements the Dynamic Covariance Scaling robust error model
* from the paper Robust Map Optimization (Agarwal13icra).
*
* Under the special condition of the parameter c == 1.0 and not
* forcing the output weight s <= 1.0, DCS is similar to Geman-McClure.
*
* This model has a scalar parameter "c" (with "units" of squared error).
*
* - Loss \rho(x) = (c²x² + cx⁴)/(x²+c)² (for any "x")
* - Derivative \phi(x) = 2c²x/(x²+c)²
* - Weight w(x) = \phi(x)/x = 2c²/(x²+c)² if x²>c, 1 otherwise
*/
class GTSAM_EXPORT DCS : public Base {
public:
typedef std::shared_ptr<DCS> shared_ptr;
DCS(double c = 1.0, const ReweightScheme reweight = Block);
~DCS() override {}
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return c_; }
protected:
double c_;
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(c_);
}
};
/** L2WithDeadZone implements a standard L2 penalty, but with a dead zone of
* width 2*k, centered at the origin. The resulting penalty within the dead
* zone is always zero, and grows quadratically outside the dead zone. In this
* sense, the L2WithDeadZone penalty is "robust to inliers", rather than being
* robust to outliers. This penalty can be used to create barrier functions in
* a general way.
*
* This model has a scalar parameter "k".
*
* - Loss \rho(x) = 0 if |x|<k, 0.5(k-|x|)² otherwise
* - Derivative \phi(x) = 0 if |x|<k, (-k+x) if x>k, (k+x) if x<-k
* - Weight w(x) = \phi(x)/x = 0 if |x|<k, (-k+x)/x if x>k, (k+x)/x if x<-k
*/
class GTSAM_EXPORT L2WithDeadZone : public Base {
protected:
double k_;
public:
typedef std::shared_ptr<L2WithDeadZone> shared_ptr;
L2WithDeadZone(double k = 1.0, const ReweightScheme reweight = Block);
double weight(double distance) const override;
double loss(double distance) const override;
void print(const std::string &s) const override;
bool equals(const Base &expected, double tol = 1e-8) const override;
static shared_ptr Create(double k, const ReweightScheme reweight = Block);
double modelParameter() const { return k_; }
private:
/** Serialization function */
friend class boost::serialization::access;
template <class ARCHIVE>
void serialize(ARCHIVE &ar, const unsigned int /*version*/) {
ar &BOOST_SERIALIZATION_BASE_OBJECT_NVP(Base);
ar &BOOST_SERIALIZATION_NVP(k_);
}
};
} // namespace mEstimator
} // namespace noiseModel
} // namespace gtsam