1 // Ceres Solver - A fast non-linear least squares minimizer 2 // Copyright 2015 Google Inc. All rights reserved. 3 // http://ceres-solver.org/ 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 // For generalized bi-partite Jacobian matrices that arise in 32 // Structure from Motion related problems, it is sometimes useful to 33 // have access to the two parts of the matrix as linear operators 34 // themselves. This class provides that functionality. 35 36 #ifndef CERES_INTERNAL_PARTITIONED_MATRIX_VIEW_H_ 37 #define CERES_INTERNAL_PARTITIONED_MATRIX_VIEW_H_ 38 39 #include <algorithm> 40 #include <cstring> 41 #include <vector> 42 43 #include "ceres/block_structure.h" 44 #include "ceres/internal/eigen.h" 45 #include "ceres/internal/port.h" 46 #include "ceres/linear_solver.h" 47 #include "ceres/small_blas.h" 48 #include "glog/logging.h" 49 50 namespace ceres { 51 namespace internal { 52 53 // Given generalized bi-partite matrix A = [E F], with the same block 54 // structure as required by the Schur complement based solver, found 55 // in explicit_schur_complement_solver.h, provide access to the 56 // matrices E and F and their outer products E'E and F'F with 57 // themselves. 58 // 59 // Lack of BlockStructure object will result in a crash and if the 60 // block structure of the matrix does not satisfy the requirements of 61 // the Schur complement solver it will result in unpredictable and 62 // wrong output. 63 class CERES_EXPORT_INTERNAL PartitionedMatrixViewBase { 64 public: ~PartitionedMatrixViewBase()65 virtual ~PartitionedMatrixViewBase() {} 66 67 // y += E'x 68 virtual void LeftMultiplyE(const double* x, double* y) const = 0; 69 70 // y += F'x 71 virtual void LeftMultiplyF(const double* x, double* y) const = 0; 72 73 // y += Ex 74 virtual void RightMultiplyE(const double* x, double* y) const = 0; 75 76 // y += Fx 77 virtual void RightMultiplyF(const double* x, double* y) const = 0; 78 79 // Create and return the block diagonal of the matrix E'E. 80 virtual BlockSparseMatrix* CreateBlockDiagonalEtE() const = 0; 81 82 // Create and return the block diagonal of the matrix F'F. Caller 83 // owns the result. 84 virtual BlockSparseMatrix* CreateBlockDiagonalFtF() const = 0; 85 86 // Compute the block diagonal of the matrix E'E and store it in 87 // block_diagonal. The matrix block_diagonal is expected to have a 88 // BlockStructure (preferably created using 89 // CreateBlockDiagonalMatrixEtE) which is has the same structure as 90 // the block diagonal of E'E. 91 virtual void UpdateBlockDiagonalEtE( 92 BlockSparseMatrix* block_diagonal) const = 0; 93 94 // Compute the block diagonal of the matrix F'F and store it in 95 // block_diagonal. The matrix block_diagonal is expected to have a 96 // BlockStructure (preferably created using 97 // CreateBlockDiagonalMatrixFtF) which is has the same structure as 98 // the block diagonal of F'F. 99 virtual void UpdateBlockDiagonalFtF( 100 BlockSparseMatrix* block_diagonal) const = 0; 101 102 // clang-format off 103 virtual int num_col_blocks_e() const = 0; 104 virtual int num_col_blocks_f() const = 0; 105 virtual int num_cols_e() const = 0; 106 virtual int num_cols_f() const = 0; 107 virtual int num_rows() const = 0; 108 virtual int num_cols() const = 0; 109 // clang-format on 110 111 static PartitionedMatrixViewBase* Create(const LinearSolver::Options& options, 112 const BlockSparseMatrix& matrix); 113 }; 114 115 template <int kRowBlockSize = Eigen::Dynamic, 116 int kEBlockSize = Eigen::Dynamic, 117 int kFBlockSize = Eigen::Dynamic> 118 class PartitionedMatrixView : public PartitionedMatrixViewBase { 119 public: 120 // matrix = [E F], where the matrix E contains the first 121 // num_col_blocks_a column blocks. 122 PartitionedMatrixView(const BlockSparseMatrix& matrix, int num_col_blocks_e); 123 124 virtual ~PartitionedMatrixView(); 125 void LeftMultiplyE(const double* x, double* y) const final; 126 void LeftMultiplyF(const double* x, double* y) const final; 127 void RightMultiplyE(const double* x, double* y) const final; 128 void RightMultiplyF(const double* x, double* y) const final; 129 BlockSparseMatrix* CreateBlockDiagonalEtE() const final; 130 BlockSparseMatrix* CreateBlockDiagonalFtF() const final; 131 void UpdateBlockDiagonalEtE(BlockSparseMatrix* block_diagonal) const final; 132 void UpdateBlockDiagonalFtF(BlockSparseMatrix* block_diagonal) const final; 133 // clang-format off num_col_blocks_e()134 int num_col_blocks_e() const final { return num_col_blocks_e_; } num_col_blocks_f()135 int num_col_blocks_f() const final { return num_col_blocks_f_; } num_cols_e()136 int num_cols_e() const final { return num_cols_e_; } num_cols_f()137 int num_cols_f() const final { return num_cols_f_; } num_rows()138 int num_rows() const final { return matrix_.num_rows(); } num_cols()139 int num_cols() const final { return matrix_.num_cols(); } 140 // clang-format on 141 142 private: 143 BlockSparseMatrix* CreateBlockDiagonalMatrixLayout(int start_col_block, 144 int end_col_block) const; 145 146 const BlockSparseMatrix& matrix_; 147 int num_row_blocks_e_; 148 int num_col_blocks_e_; 149 int num_col_blocks_f_; 150 int num_cols_e_; 151 int num_cols_f_; 152 }; 153 154 } // namespace internal 155 } // namespace ceres 156 157 #endif // CERES_INTERNAL_PARTITIONED_MATRIX_VIEW_H_ 158