ceres中的ConditionedCostFuntion源码阅读

ceres中的ConditionedCostFuntion源码阅读

class ConditionedCostFuntion
这个类的作用是给隐藏代价函数（cost function）的残差值赋上不同的权值。举个简单的例子：你有一个产生了N个残差值的代价函数，但是你想要残差的总体值不是简单的等于每个参加的平方和，你想给每一个残差一个特定的权值（尺度），来改变每个残差对总体参加的贡献。

//  my_cost_function produces N residuals 
CostFunction* my_cost_function = ... 
CHECK_EQ(N, my_cost_function->num_residuals()); 
vector<CostFunction*> conditioners; 
 
//  Make N 1x1 cost functions (1 parameter, 1 residual) 
CostFunction* f_1 = ... 
conditioners.push_back(f_1); 
 
CostFunction* f_N = ... conditioners.push_back(f_N); 
ConditionedCostFunction* ccf =   
new ConditionedCostFunction(my_cost_function, conditioners);

这样，ccf的residual[i]可以从第i个conditioner中传出

ccf_residual[i] = f_i(my_cost_function_residual[i])

相应的，Jacobian会被合理的影响。
下面贴上源码，头文件，CPP及官方test
一目了然，正如遇事不决，读源码

//
// Author: wjr@google.com (William Rucklidge)
//
// This file contains a cost function that can apply a transformation to
// each residual value before they are square-summed.

#ifndef CERES_PUBLIC_CONDITIONED_COST_FUNCTION_H_
#define CERES_PUBLIC_CONDITIONED_COST_FUNCTION_H_

#include <memory>
#include <vector>

#include "ceres/cost_function.h"
#include "ceres/internal/disable_warnings.h"
#include "ceres/types.h"

namespace ceres {

// This class allows you to apply different conditioning to the residual
// values of a wrapped cost function. An example where this is useful is
// where you have an existing cost function that produces N values, but you
// want the total cost to be something other than just the sum of these
// squared values - maybe you want to apply a different scaling to some
// values, to change their contribution to the cost.
//
// Usage:
//
//   // my_cost_function produces N residuals
//   CostFunction* my_cost_function = ...
//   CHECK_EQ(N, my_cost_function->num_residuals());
//   vector<CostFunction*> conditioners;
//
//   // Make N 1x1 cost functions (1 parameter, 1 residual)
//   CostFunction* f_1 = ...
//   conditioners.push_back(f_1);
//   ...
//   CostFunction* f_N = ...
//   conditioners.push_back(f_N);
//   ConditionedCostFunction* ccf =
//     new ConditionedCostFunction(my_cost_function, conditioners);
//
// Now ccf's residual i (i=0..N-1) will be passed though the i'th conditioner.
//
//   ccf_residual[i] = f_i(my_cost_function_residual[i])
//
// and the Jacobian will be affected appropriately.
class CERES_EXPORT ConditionedCostFunction : public CostFunction {
 public:
  // Builds a cost function based on a wrapped cost function, and a
  // per-residual conditioner. Takes ownership of all of the wrapped cost
  // functions, or not, depending on the ownership parameter. Conditioners
  // may be NULL, in which case the corresponding residual is not modified.
  //
  // The conditioners can repeat.
  ConditionedCostFunction(CostFunction* wrapped_cost_function,
                          const std::vector<CostFunction*>& conditioners,
                          Ownership ownership);
  virtual ~ConditionedCostFunction();

  bool Evaluate(double const* const* parameters,
                double* residuals,
                double** jacobians) const override;

 private:
  std::unique_ptr<CostFunction> wrapped_cost_function_;
  std::vector<CostFunction*> conditioners_;
  Ownership ownership_;
};

}  // namespace ceres

#include "ceres/internal/reenable_warnings.h"

#endif  // CERES_PUBLIC_CONDITIONED_COST_FUNCTION_H_

// Ceres Solver - A fast non-linear least squares minimizer
// Copyright 2015 Google Inc. All rights reserved.
// http://ceres-solver.org/
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// * Redistributions of source code must retain the above copyright notice,
//   this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above copyright notice,
//   this list of conditions and the following disclaimer in the documentation
//   and/or other materials provided with the distribution.
// * Neither the name of Google Inc. nor the names of its contributors may be
//   used to endorse or promote products derived from this software without
//   specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
// Author: wjr@google.com (William Rucklidge)
//
// Tests for the conditioned cost function.

#include "ceres/conditioned_cost_function.h"

#include "ceres/internal/eigen.h"
#include "ceres/normal_prior.h"
#include "ceres/types.h"
#include "gtest/gtest.h"

namespace ceres {
namespace internal {

// The size of the cost functions we build.
static const int kTestCostFunctionSize = 3;

// A simple cost function: return ax + b.
class LinearCostFunction : public CostFunction {
 public:
  LinearCostFunction(double a, double b) : a_(a), b_(b) {
    set_num_residuals(1);
    mutable_parameter_block_sizes()->push_back(1);
  }

  bool Evaluate(double const* const* parameters,
                double* residuals,
                double** jacobians) const final {
    *residuals = **parameters * a_ + b_;
    if (jacobians && *jacobians) {
      **jacobians = a_;
    }

    return true;
  }

 private:
  const double a_, b_;
};

// Tests that ConditionedCostFunction does what it's supposed to.
TEST(ConditionedCostFunction, NormalOperation) {
  double v1[kTestCostFunctionSize], v2[kTestCostFunctionSize],
      jac[kTestCostFunctionSize * kTestCostFunctionSize],
      result[kTestCostFunctionSize];

  for (int i = 0; i < kTestCostFunctionSize; i++) {
    v1[i] = i;
    v2[i] = i * 10;
    // Seed a few garbage values in the Jacobian matrix, to make sure that
    // they're overwritten.
    jac[i * 2] = i * i;
    result[i] = i * i * i;
  }

  // Make a cost function that computes x - v2
  VectorRef v2_vector(v2, kTestCostFunctionSize, 1);
  Matrix identity(kTestCostFunctionSize, kTestCostFunctionSize);
  identity.setIdentity();
  NormalPrior* difference_cost_function = new NormalPrior(identity, v2_vector);

  std::vector<CostFunction*> conditioners;
  for (int i = 0; i < kTestCostFunctionSize; i++) {
    conditioners.push_back(new LinearCostFunction(i + 2, i * 7));
  }

  ConditionedCostFunction conditioned_cost_function(
      difference_cost_function, conditioners, TAKE_OWNERSHIP);
  EXPECT_EQ(difference_cost_function->num_residuals(),
            conditioned_cost_function.num_residuals());
  EXPECT_EQ(difference_cost_function->parameter_block_sizes(),
            conditioned_cost_function.parameter_block_sizes());

  double* parameters[1];
  parameters[0] = v1;
  double* jacs[1];
  jacs[0] = jac;

  conditioned_cost_function.Evaluate(parameters, result, jacs);
  for (int i = 0; i < kTestCostFunctionSize; i++) {
    EXPECT_DOUBLE_EQ((i + 2) * (v1[i] - v2[i]) + i * 7, result[i]);
  }

  for (int i = 0; i < kTestCostFunctionSize; i++) {
    for (int j = 0; j < kTestCostFunctionSize; j++) {
      double actual = jac[i * kTestCostFunctionSize + j];
      if (i != j) {
        EXPECT_DOUBLE_EQ(0, actual);
      } else {
        EXPECT_DOUBLE_EQ(i + 2, actual);
      }
    }
  }
}

TEST(ConditionedCostFunction, SharedConditionersDoNotTriggerDoubleFree) {
  // Make a cost function that computes x - v2
  double v2[kTestCostFunctionSize];
  VectorRef v2_vector(v2, kTestCostFunctionSize, 1);
  Matrix identity = Matrix::Identity(kTestCostFunctionSize, kTestCostFunctionSize);
  NormalPrior* difference_cost_function = new NormalPrior(identity, v2_vector);
  CostFunction* conditioner = new LinearCostFunction(2, 7);
  std::vector<CostFunction*> conditioners;
  for (int i = 0; i < kTestCostFunctionSize; i++) {
    conditioners.push_back(conditioner);
  }

  ConditionedCostFunction conditioned_cost_function(
      difference_cost_function, conditioners, TAKE_OWNERSHIP);
}

}  // namespace internal
}  // namespace ceres

// Ceres Solver - A fast non-linear least squares minimizer
// Copyright 2015 Google Inc. All rights reserved.
// http://ceres-solver.org/
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// * Redistributions of source code must retain the above copyright notice,
//   this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above copyright notice,
//   this list of conditions and the following disclaimer in the documentation
//   and/or other materials provided with the distribution.
// * Neither the name of Google Inc. nor the names of its contributors may be
//   used to endorse or promote products derived from this software without
//   specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
// Author: wjr@google.com (William Rucklidge)
//
// This file contains the implementation of the conditioned cost function.

#include "ceres/conditioned_cost_function.h"

#include <cstddef>

#include "ceres/internal/eigen.h"
#include "ceres/stl_util.h"
#include "ceres/types.h"
#include "glog/logging.h"

namespace ceres {

// This cost function has the same dimensions (parameters, residuals) as
// the one it's wrapping.
ConditionedCostFunction::ConditionedCostFunction(
    CostFunction* wrapped_cost_function,
    const std::vector<CostFunction*>& conditioners,
    Ownership ownership)
    : wrapped_cost_function_(wrapped_cost_function),
      conditioners_(conditioners),
      ownership_(ownership) {
  // Set up our dimensions.
  set_num_residuals(wrapped_cost_function_->num_residuals());
  *mutable_parameter_block_sizes() =
      wrapped_cost_function_->parameter_block_sizes();

  // Sanity-check the conditioners' dimensions.
  CHECK_EQ(wrapped_cost_function_->num_residuals(), conditioners_.size());
  for (int i = 0; i < wrapped_cost_function_->num_residuals(); i++) {
    if (conditioners[i]) {
      CHECK_EQ(1, conditioners[i]->num_residuals());
      CHECK_EQ(1, conditioners[i]->parameter_block_sizes().size());
      CHECK_EQ(1, conditioners[i]->parameter_block_sizes()[0]);
    }
  }
}

ConditionedCostFunction::~ConditionedCostFunction() {
  if (ownership_ == TAKE_OWNERSHIP) {
    STLDeleteUniqueContainerPointers(conditioners_.begin(), conditioners_.end());
  } else {
    wrapped_cost_function_.release();
  }
}

bool ConditionedCostFunction::Evaluate(double const* const* parameters,
                                       double* residuals,
                                       double** jacobians) const {
  bool success = wrapped_cost_function_->Evaluate(parameters, residuals,
                                                  jacobians);
  if (!success) {
    return false;
  }

  for (int r = 0; r < wrapped_cost_function_->num_residuals(); r++) {
    // On output, we want to have
    // residuals[r] = conditioners[r](wrapped_residuals[r])
    // For parameter block i, column c,
    // jacobians[i][r*parameter_block_size_[i] + c] =
    //   = d residual[r] / d parameters[i][c]
    //   = conditioners[r]'(wrapped_residuals[r]) *
    //       d wrapped_residuals[r] / d parameters[i][c]
    if (conditioners_[r]) {
      double conditioner_derivative;
      double* conditioner_derivative_pointer = &conditioner_derivative;
      double** conditioner_derivative_pointer2 =
          &conditioner_derivative_pointer;
      if (!jacobians) {
        conditioner_derivative_pointer2 = NULL;
      }

      double unconditioned_residual = residuals[r];
      double* parameter_pointer = &unconditioned_residual;
      success = conditioners_[r]->Evaluate(&parameter_pointer,
                                           &residuals[r],
                                           conditioner_derivative_pointer2);
      if (!success) {
        return false;
      }

      if (jacobians) {
        for (int i = 0;
             i < wrapped_cost_function_->parameter_block_sizes().size();
             i++) {
          if (jacobians[i]) {
            int parameter_block_size =
                wrapped_cost_function_->parameter_block_sizes()[i];
            VectorRef jacobian_row(jacobians[i] + r * parameter_block_size,
                                   parameter_block_size, 1);
            jacobian_row *= conditioner_derivative;
          }
        }
      }
    }
  }
  return true;
}

}  // namespace ceres