• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #ifndef CERES_INTERNAL_SCHUR_COMPLEMENT_SOLVER_H_
32 #define CERES_INTERNAL_SCHUR_COMPLEMENT_SOLVER_H_
33 
34 #include <set>
35 #include <utility>
36 #include <vector>
37 
38 #include "ceres/internal/port.h"
39 
40 #include "ceres/block_random_access_matrix.h"
41 #include "ceres/block_sparse_matrix.h"
42 #include "ceres/block_structure.h"
43 #include "ceres/cxsparse.h"
44 #include "ceres/linear_solver.h"
45 #include "ceres/schur_eliminator.h"
46 #include "ceres/suitesparse.h"
47 #include "ceres/internal/scoped_ptr.h"
48 #include "ceres/types.h"
49 
50 #ifdef CERES_USE_EIGEN_SPARSE
51 #include "Eigen/SparseCholesky"
52 #endif
53 
54 namespace ceres {
55 namespace internal {
56 
57 class BlockSparseMatrix;
58 
59 // Base class for Schur complement based linear least squares
60 // solvers. It assumes that the input linear system Ax = b can be
61 // partitioned into
62 //
63 //  E y + F z = b
64 //
65 // Where x = [y;z] is a partition of the variables.  The paritioning
66 // of the variables is such that, E'E is a block diagonal
67 // matrix. Further, the rows of A are ordered so that for every
68 // variable block in y, all the rows containing that variable block
69 // occur as a vertically contiguous block. i.e the matrix A looks like
70 //
71 //              E                 F
72 //  A = [ y1   0   0   0 |  z1    0    0   0    z5]
73 //      [ y1   0   0   0 |  z1   z2    0   0     0]
74 //      [  0  y2   0   0 |   0    0   z3   0     0]
75 //      [  0   0  y3   0 |  z1   z2   z3  z4    z5]
76 //      [  0   0  y3   0 |  z1    0    0   0    z5]
77 //      [  0   0   0  y4 |   0    0    0   0    z5]
78 //      [  0   0   0  y4 |   0   z2    0   0     0]
79 //      [  0   0   0  y4 |   0    0    0   0     0]
80 //      [  0   0   0   0 |  z1    0    0   0     0]
81 //      [  0   0   0   0 |   0    0   z3  z4    z5]
82 //
83 // This structure should be reflected in the corresponding
84 // CompressedRowBlockStructure object associated with A. The linear
85 // system Ax = b should either be well posed or the array D below
86 // should be non-null and the diagonal matrix corresponding to it
87 // should be non-singular.
88 //
89 // SchurComplementSolver has two sub-classes.
90 //
91 // DenseSchurComplementSolver: For problems where the Schur complement
92 // matrix is small and dense, or if CHOLMOD/SuiteSparse is not
93 // installed. For structure from motion problems, this is solver can
94 // be used for problems with upto a few hundred cameras.
95 //
96 // SparseSchurComplementSolver: For problems where the Schur
97 // complement matrix is large and sparse. It requires that
98 // CHOLMOD/SuiteSparse be installed, as it uses CHOLMOD to find a
99 // sparse Cholesky factorization of the Schur complement. This solver
100 // can be used for solving structure from motion problems with tens of
101 // thousands of cameras, though depending on the exact sparsity
102 // structure, it maybe better to use an iterative solver.
103 //
104 // The two solvers can be instantiated by calling
105 // LinearSolver::CreateLinearSolver with LinearSolver::Options::type
106 // set to DENSE_SCHUR and SPARSE_SCHUR
107 // respectively. LinearSolver::Options::elimination_groups[0] should be
108 // at least 1.
109 class SchurComplementSolver : public BlockSparseMatrixSolver {
110  public:
SchurComplementSolver(const LinearSolver::Options & options)111   explicit SchurComplementSolver(const LinearSolver::Options& options)
112       : options_(options) {
113     CHECK_GT(options.elimination_groups.size(), 1);
114     CHECK_GT(options.elimination_groups[0], 0);
115   }
116 
117   // LinearSolver methods
~SchurComplementSolver()118   virtual ~SchurComplementSolver() {}
119   virtual LinearSolver::Summary SolveImpl(
120       BlockSparseMatrix* A,
121       const double* b,
122       const LinearSolver::PerSolveOptions& per_solve_options,
123       double* x);
124 
125  protected:
options()126   const LinearSolver::Options& options() const { return options_; }
127 
lhs()128   const BlockRandomAccessMatrix* lhs() const { return lhs_.get(); }
set_lhs(BlockRandomAccessMatrix * lhs)129   void set_lhs(BlockRandomAccessMatrix* lhs) { lhs_.reset(lhs); }
rhs()130   const double* rhs() const { return rhs_.get(); }
set_rhs(double * rhs)131   void set_rhs(double* rhs) { rhs_.reset(rhs); }
132 
133  private:
134   virtual void InitStorage(const CompressedRowBlockStructure* bs) = 0;
135   virtual LinearSolver::Summary SolveReducedLinearSystem(
136       double* solution) = 0;
137 
138   LinearSolver::Options options_;
139 
140   scoped_ptr<SchurEliminatorBase> eliminator_;
141   scoped_ptr<BlockRandomAccessMatrix> lhs_;
142   scoped_array<double> rhs_;
143 
144   CERES_DISALLOW_COPY_AND_ASSIGN(SchurComplementSolver);
145 };
146 
147 // Dense Cholesky factorization based solver.
148 class DenseSchurComplementSolver : public SchurComplementSolver {
149  public:
DenseSchurComplementSolver(const LinearSolver::Options & options)150   explicit DenseSchurComplementSolver(const LinearSolver::Options& options)
151       : SchurComplementSolver(options) {}
~DenseSchurComplementSolver()152   virtual ~DenseSchurComplementSolver() {}
153 
154  private:
155   virtual void InitStorage(const CompressedRowBlockStructure* bs);
156   virtual LinearSolver::Summary SolveReducedLinearSystem(
157       double* solution);
158 
159   CERES_DISALLOW_COPY_AND_ASSIGN(DenseSchurComplementSolver);
160 };
161 
162 // Sparse Cholesky factorization based solver.
163 class SparseSchurComplementSolver : public SchurComplementSolver {
164  public:
165   explicit SparseSchurComplementSolver(const LinearSolver::Options& options);
166   virtual ~SparseSchurComplementSolver();
167 
168  private:
169   virtual void InitStorage(const CompressedRowBlockStructure* bs);
170   virtual LinearSolver::Summary SolveReducedLinearSystem(
171       double* solution);
172   LinearSolver::Summary SolveReducedLinearSystemUsingSuiteSparse(
173       double* solution);
174   LinearSolver::Summary SolveReducedLinearSystemUsingCXSparse(
175       double* solution);
176   LinearSolver::Summary SolveReducedLinearSystemUsingEigen(
177       double* solution);
178 
179   // Size of the blocks in the Schur complement.
180   vector<int> blocks_;
181 
182   SuiteSparse ss_;
183   // Symbolic factorization of the reduced linear system. Precomputed
184   // once and reused in subsequent calls.
185   cholmod_factor* factor_;
186 
187   CXSparse cxsparse_;
188   // Cached factorization
189   cs_dis* cxsparse_factor_;
190 
191 #ifdef CERES_USE_EIGEN_SPARSE
192   typedef Eigen::SimplicialLDLT<Eigen::SparseMatrix<double> > SimplicialLDLT;
193   scoped_ptr<SimplicialLDLT> simplicial_ldlt_;
194 #endif
195 
196   CERES_DISALLOW_COPY_AND_ASSIGN(SparseSchurComplementSolver);
197 };
198 
199 }  // namespace internal
200 }  // namespace ceres
201 
202 #endif  // CERES_INTERNAL_SCHUR_COMPLEMENT_SOLVER_H_
203