1 // Ceres Solver - A fast non-linear least squares minimizer
2 // Copyright 2010, 2011, 2012 Google Inc. All rights reserved.
3 // http://code.google.com/p/ceres-solver/
4 //
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are met:
7 //
8 // * Redistributions of source code must retain the above copyright notice,
9 //   this list of conditions and the following disclaimer.
10 // * Redistributions in binary form must reproduce the above copyright notice,
11 //   this list of conditions and the following disclaimer in the documentation
12 //   and/or other materials provided with the distribution.
13 // * Neither the name of Google Inc. nor the names of its contributors may be
14 //   used to endorse or promote products derived from this software without
15 //   specific prior written permission.
16 //
17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 // POSSIBILITY OF SUCH DAMAGE.
28 //
29 // Author: sameeragarwal@google.com (Sameer Agarwal)
30 //
31 // Create CostFunctions as needed by the least squares framework, with
32 // Jacobians computed via automatic differentiation. For more
33 // information on automatic differentation, see the wikipedia article
34 // at http://en.wikipedia.org/wiki/Automatic_differentiation
35 //
36 // To get an auto differentiated cost function, you must define a class with a
37 // templated operator() (a functor) that computes the cost function in terms of
38 // the template parameter T. The autodiff framework substitutes appropriate
39 // "jet" objects for T in order to compute the derivative when necessary, but
40 // this is hidden, and you should write the function as if T were a scalar type
41 // (e.g. a double-precision floating point number).
42 //
43 // The function must write the computed value in the last argument
44 // (the only non-const one) and return true to indicate
45 // success. Please see cost_function.h for details on how the return
46 // value maybe used to impose simple constraints on the parameter
47 // block.
48 //
49 // For example, consider a scalar error e = k - x'y, where both x and y are
50 // two-dimensional column vector parameters, the prime sign indicates
51 // transposition, and k is a constant. The form of this error, which is the
52 // difference between a constant and an expression, is a common pattern in least
53 // squares problems. For example, the value x'y might be the model expectation
54 // for a series of measurements, where there is an instance of the cost function
55 // for each measurement k.
56 //
57 // The actual cost added to the total problem is e^2, or (k - x'k)^2; however,
58 // the squaring is implicitly done by the optimization framework.
59 //
60 // To write an auto-differentiable cost function for the above model, first
61 // define the object
62 //
63 //   class MyScalarCostFunctor {
64 //     MyScalarCostFunctor(double k): k_(k) {}
65 //
66 //     template <typename T>
67 //     bool operator()(const T* const x , const T* const y, T* e) const {
68 //       e[0] = T(k_) - x[0] * y[0] + x[1] * y[1];
69 //       return true;
70 //     }
71 //
72 //    private:
73 //     double k_;
74 //   };
75 //
76 // Note that in the declaration of operator() the input parameters x and y come
77 // first, and are passed as const pointers to arrays of T. If there were three
78 // input parameters, then the third input parameter would come after y. The
79 // output is always the last parameter, and is also a pointer to an array. In
80 // the example above, e is a scalar, so only e[0] is set.
81 //
82 // Then given this class definition, the auto differentiated cost function for
83 // it can be constructed as follows.
84 //
85 //   CostFunction* cost_function
86 //       = new AutoDiffCostFunction<MyScalarCostFunctor, 1, 2, 2>(
87 //            new MyScalarCostFunctor(1.0));             ^  ^  ^
88 //                                                       |  |  |
89 //                            Dimension of residual -----+  |  |
90 //                            Dimension of x ---------------+  |
91 //                            Dimension of y ------------------+
92 //
93 // In this example, there is usually an instance for each measumerent of k.
94 //
95 // In the instantiation above, the template parameters following
96 // "MyScalarCostFunctor", "1, 2, 2", describe the functor as computing a
97 // 1-dimensional output from two arguments, both 2-dimensional.
98 //
99 // AutoDiffCostFunction also supports cost functions with a
100 // runtime-determined number of residuals. For example:
101 //
102 //   CostFunction* cost_function
103 //       = new AutoDiffCostFunction<MyScalarCostFunctor, DYNAMIC, 2, 2>(
104 //           new CostFunctorWithDynamicNumResiduals(1.0),   ^     ^  ^
105 //           runtime_number_of_residuals); <----+           |     |  |
106 //                                              |           |     |  |
107 //                                              |           |     |  |
108 //             Actual number of residuals ------+           |     |  |
109 //             Indicate dynamic number of residuals --------+     |  |
110 //             Dimension of x ------------------------------------+  |
111 //             Dimension of y ---------------------------------------+
112 //
113 // The framework can currently accommodate cost functions of up to 10
114 // independent variables, and there is no limit on the dimensionality
115 // of each of them.
116 //
117 // WARNING #1: Since the functor will get instantiated with different types for
118 // T, you must to convert from other numeric types to T before mixing
119 // computations with other variables of type T. In the example above, this is
120 // seen where instead of using k_ directly, k_ is wrapped with T(k_).
121 //
122 // WARNING #2: A common beginner's error when first using autodiff cost
123 // functions is to get the sizing wrong. In particular, there is a tendency to
124 // set the template parameters to (dimension of residual, number of parameters)
125 // instead of passing a dimension parameter for *every parameter*. In the
126 // example above, that would be <MyScalarCostFunctor, 1, 2>, which is missing
127 // the last '2' argument. Please be careful when setting the size parameters.
128 
129 #ifndef CERES_PUBLIC_AUTODIFF_COST_FUNCTION_H_
130 #define CERES_PUBLIC_AUTODIFF_COST_FUNCTION_H_
131 
132 #include "ceres/internal/autodiff.h"
133 #include "ceres/internal/scoped_ptr.h"
134 #include "ceres/sized_cost_function.h"
135 #include "ceres/types.h"
136 #include "glog/logging.h"
137 
138 namespace ceres {
139 
140 // A cost function which computes the derivative of the cost with respect to
141 // the parameters (a.k.a. the jacobian) using an autodifferentiation framework.
142 // The first template argument is the functor object, described in the header
143 // comment. The second argument is the dimension of the residual (or
144 // ceres::DYNAMIC to indicate it will be set at runtime), and subsequent
145 // arguments describe the size of the Nth parameter, one per parameter.
146 //
147 // The constructors take ownership of the cost functor.
148 //
149 // If the number of residuals (argument kNumResiduals below) is
150 // ceres::DYNAMIC, then the two-argument constructor must be used. The
151 // second constructor takes a number of residuals (in addition to the
152 // templated number of residuals). This allows for varying the number
153 // of residuals for a single autodiff cost function at runtime.
154 template <typename CostFunctor,
155           int kNumResiduals,  // Number of residuals, or ceres::DYNAMIC.
156           int N0,       // Number of parameters in block 0.
157           int N1 = 0,   // Number of parameters in block 1.
158           int N2 = 0,   // Number of parameters in block 2.
159           int N3 = 0,   // Number of parameters in block 3.
160           int N4 = 0,   // Number of parameters in block 4.
161           int N5 = 0,   // Number of parameters in block 5.
162           int N6 = 0,   // Number of parameters in block 6.
163           int N7 = 0,   // Number of parameters in block 7.
164           int N8 = 0,   // Number of parameters in block 8.
165           int N9 = 0>   // Number of parameters in block 9.
166 class AutoDiffCostFunction : public SizedCostFunction<kNumResiduals,
167                                                       N0, N1, N2, N3, N4,
168                                                       N5, N6, N7, N8, N9> {
169  public:
170   // Takes ownership of functor. Uses the template-provided value for the
171   // number of residuals ("kNumResiduals").
AutoDiffCostFunction(CostFunctor * functor)172   explicit AutoDiffCostFunction(CostFunctor* functor)
173       : functor_(functor) {
174     CHECK_NE(kNumResiduals, DYNAMIC)
175         << "Can't run the fixed-size constructor if the "
176         << "number of residuals is set to ceres::DYNAMIC.";
177   }
178 
179   // Takes ownership of functor. Ignores the template-provided
180   // kNumResiduals in favor of the "num_residuals" argument provided.
181   //
182   // This allows for having autodiff cost functions which return varying
183   // numbers of residuals at runtime.
AutoDiffCostFunction(CostFunctor * functor,int num_residuals)184   AutoDiffCostFunction(CostFunctor* functor, int num_residuals)
185       : functor_(functor) {
186     CHECK_EQ(kNumResiduals, DYNAMIC)
187         << "Can't run the dynamic-size constructor if the "
188         << "number of residuals is not ceres::DYNAMIC.";
189     SizedCostFunction<kNumResiduals,
190                       N0, N1, N2, N3, N4,
191                       N5, N6, N7, N8, N9>
192         ::set_num_residuals(num_residuals);
193   }
194 
~AutoDiffCostFunction()195   virtual ~AutoDiffCostFunction() {}
196 
197   // Implementation details follow; clients of the autodiff cost function should
198   // not have to examine below here.
199   //
200   // To handle varardic cost functions, some template magic is needed. It's
201   // mostly hidden inside autodiff.h.
Evaluate(double const * const * parameters,double * residuals,double ** jacobians)202   virtual bool Evaluate(double const* const* parameters,
203                         double* residuals,
204                         double** jacobians) const {
205     if (!jacobians) {
206       return internal::VariadicEvaluate<
207           CostFunctor, double, N0, N1, N2, N3, N4, N5, N6, N7, N8, N9>
208           ::Call(*functor_, parameters, residuals);
209     }
210     return internal::AutoDiff<CostFunctor, double,
211            N0, N1, N2, N3, N4, N5, N6, N7, N8, N9>::Differentiate(
212                *functor_,
213                parameters,
214                SizedCostFunction<kNumResiduals,
215                                  N0, N1, N2, N3, N4,
216                                  N5, N6, N7, N8, N9>::num_residuals(),
217                residuals,
218                jacobians);
219   }
220 
221  private:
222   internal::scoped_ptr<CostFunctor> functor_;
223 };
224 
225 }  // namespace ceres
226 
227 #endif  // CERES_PUBLIC_AUTODIFF_COST_FUNCTION_H_
228