Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 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 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 31 | #include "ceres/local_parameterization.h" |
| 32 | |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 33 | #include <cmath> |
| 34 | #include <limits> |
| 35 | #include <memory> |
| 36 | |
| 37 | #include "Eigen/Geometry" |
| 38 | #include "ceres/autodiff_local_parameterization.h" |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 39 | #include "ceres/internal/autodiff.h" |
| 40 | #include "ceres/internal/eigen.h" |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 41 | #include "ceres/internal/householder_vector.h" |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 42 | #include "ceres/random.h" |
| 43 | #include "ceres/rotation.h" |
| 44 | #include "gtest/gtest.h" |
| 45 | |
| 46 | namespace ceres { |
| 47 | namespace internal { |
| 48 | |
| 49 | TEST(IdentityParameterization, EverythingTest) { |
| 50 | IdentityParameterization parameterization(3); |
| 51 | EXPECT_EQ(parameterization.GlobalSize(), 3); |
| 52 | EXPECT_EQ(parameterization.LocalSize(), 3); |
| 53 | |
| 54 | double x[3] = {1.0, 2.0, 3.0}; |
| 55 | double delta[3] = {0.0, 1.0, 2.0}; |
| 56 | double x_plus_delta[3] = {0.0, 0.0, 0.0}; |
| 57 | parameterization.Plus(x, delta, x_plus_delta); |
| 58 | EXPECT_EQ(x_plus_delta[0], 1.0); |
| 59 | EXPECT_EQ(x_plus_delta[1], 3.0); |
| 60 | EXPECT_EQ(x_plus_delta[2], 5.0); |
| 61 | |
| 62 | double jacobian[9]; |
| 63 | parameterization.ComputeJacobian(x, jacobian); |
| 64 | int k = 0; |
| 65 | for (int i = 0; i < 3; ++i) { |
| 66 | for (int j = 0; j < 3; ++j, ++k) { |
| 67 | EXPECT_EQ(jacobian[k], (i == j) ? 1.0 : 0.0); |
| 68 | } |
| 69 | } |
| 70 | |
| 71 | Matrix global_matrix = Matrix::Ones(10, 3); |
| 72 | Matrix local_matrix = Matrix::Zero(10, 3); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 73 | parameterization.MultiplyByJacobian( |
| 74 | x, 10, global_matrix.data(), local_matrix.data()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 75 | EXPECT_EQ((local_matrix - global_matrix).norm(), 0.0); |
| 76 | } |
| 77 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 78 | TEST(SubsetParameterization, EmptyConstantParameters) { |
| 79 | std::vector<int> constant_parameters; |
| 80 | SubsetParameterization parameterization(3, constant_parameters); |
| 81 | EXPECT_EQ(parameterization.GlobalSize(), 3); |
| 82 | EXPECT_EQ(parameterization.LocalSize(), 3); |
| 83 | double x[3] = {1, 2, 3}; |
| 84 | double delta[3] = {4, 5, 6}; |
| 85 | double x_plus_delta[3] = {-1, -2, -3}; |
| 86 | parameterization.Plus(x, delta, x_plus_delta); |
| 87 | EXPECT_EQ(x_plus_delta[0], x[0] + delta[0]); |
| 88 | EXPECT_EQ(x_plus_delta[1], x[1] + delta[1]); |
| 89 | EXPECT_EQ(x_plus_delta[2], x[2] + delta[2]); |
| 90 | |
| 91 | Matrix jacobian(3, 3); |
| 92 | Matrix expected_jacobian(3, 3); |
| 93 | expected_jacobian.setIdentity(); |
| 94 | parameterization.ComputeJacobian(x, jacobian.data()); |
| 95 | EXPECT_EQ(jacobian, expected_jacobian); |
| 96 | |
| 97 | Matrix global_matrix(3, 5); |
| 98 | global_matrix.setRandom(); |
| 99 | Matrix local_matrix(3, 5); |
| 100 | parameterization.MultiplyByJacobian( |
| 101 | x, 5, global_matrix.data(), local_matrix.data()); |
| 102 | EXPECT_EQ(global_matrix, local_matrix); |
| 103 | } |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 104 | |
| 105 | TEST(SubsetParameterization, NegativeParameterIndexDeathTest) { |
| 106 | std::vector<int> constant_parameters; |
| 107 | constant_parameters.push_back(-1); |
| 108 | EXPECT_DEATH_IF_SUPPORTED( |
| 109 | SubsetParameterization parameterization(2, constant_parameters), |
| 110 | "greater than equal to zero"); |
| 111 | } |
| 112 | |
| 113 | TEST(SubsetParameterization, GreaterThanSizeParameterIndexDeathTest) { |
| 114 | std::vector<int> constant_parameters; |
| 115 | constant_parameters.push_back(2); |
| 116 | EXPECT_DEATH_IF_SUPPORTED( |
| 117 | SubsetParameterization parameterization(2, constant_parameters), |
| 118 | "less than the size"); |
| 119 | } |
| 120 | |
| 121 | TEST(SubsetParameterization, DuplicateParametersDeathTest) { |
| 122 | std::vector<int> constant_parameters; |
| 123 | constant_parameters.push_back(1); |
| 124 | constant_parameters.push_back(1); |
| 125 | EXPECT_DEATH_IF_SUPPORTED( |
| 126 | SubsetParameterization parameterization(2, constant_parameters), |
| 127 | "duplicates"); |
| 128 | } |
| 129 | |
| 130 | TEST(SubsetParameterization, |
| 131 | ProductParameterizationWithZeroLocalSizeSubsetParameterization1) { |
| 132 | std::vector<int> constant_parameters; |
| 133 | constant_parameters.push_back(0); |
| 134 | LocalParameterization* subset_param = |
| 135 | new SubsetParameterization(1, constant_parameters); |
| 136 | LocalParameterization* identity_param = new IdentityParameterization(2); |
| 137 | ProductParameterization product_param(subset_param, identity_param); |
| 138 | EXPECT_EQ(product_param.GlobalSize(), 3); |
| 139 | EXPECT_EQ(product_param.LocalSize(), 2); |
| 140 | double x[] = {1.0, 1.0, 1.0}; |
| 141 | double delta[] = {2.0, 3.0}; |
| 142 | double x_plus_delta[] = {0.0, 0.0, 0.0}; |
| 143 | EXPECT_TRUE(product_param.Plus(x, delta, x_plus_delta)); |
| 144 | EXPECT_EQ(x_plus_delta[0], x[0]); |
| 145 | EXPECT_EQ(x_plus_delta[1], x[1] + delta[0]); |
| 146 | EXPECT_EQ(x_plus_delta[2], x[2] + delta[1]); |
| 147 | |
| 148 | Matrix actual_jacobian(3, 2); |
| 149 | EXPECT_TRUE(product_param.ComputeJacobian(x, actual_jacobian.data())); |
| 150 | } |
| 151 | |
| 152 | TEST(SubsetParameterization, |
| 153 | ProductParameterizationWithZeroLocalSizeSubsetParameterization2) { |
| 154 | std::vector<int> constant_parameters; |
| 155 | constant_parameters.push_back(0); |
| 156 | LocalParameterization* subset_param = |
| 157 | new SubsetParameterization(1, constant_parameters); |
| 158 | LocalParameterization* identity_param = new IdentityParameterization(2); |
| 159 | ProductParameterization product_param(identity_param, subset_param); |
| 160 | EXPECT_EQ(product_param.GlobalSize(), 3); |
| 161 | EXPECT_EQ(product_param.LocalSize(), 2); |
| 162 | double x[] = {1.0, 1.0, 1.0}; |
| 163 | double delta[] = {2.0, 3.0}; |
| 164 | double x_plus_delta[] = {0.0, 0.0, 0.0}; |
| 165 | EXPECT_TRUE(product_param.Plus(x, delta, x_plus_delta)); |
| 166 | EXPECT_EQ(x_plus_delta[0], x[0] + delta[0]); |
| 167 | EXPECT_EQ(x_plus_delta[1], x[1] + delta[1]); |
| 168 | EXPECT_EQ(x_plus_delta[2], x[2]); |
| 169 | |
| 170 | Matrix actual_jacobian(3, 2); |
| 171 | EXPECT_TRUE(product_param.ComputeJacobian(x, actual_jacobian.data())); |
| 172 | } |
| 173 | |
| 174 | TEST(SubsetParameterization, NormalFunctionTest) { |
| 175 | const int kGlobalSize = 4; |
| 176 | const int kLocalSize = 3; |
| 177 | |
| 178 | double x[kGlobalSize] = {1.0, 2.0, 3.0, 4.0}; |
| 179 | for (int i = 0; i < kGlobalSize; ++i) { |
| 180 | std::vector<int> constant_parameters; |
| 181 | constant_parameters.push_back(i); |
| 182 | SubsetParameterization parameterization(kGlobalSize, constant_parameters); |
| 183 | double delta[kLocalSize] = {1.0, 2.0, 3.0}; |
| 184 | double x_plus_delta[kGlobalSize] = {0.0, 0.0, 0.0}; |
| 185 | |
| 186 | parameterization.Plus(x, delta, x_plus_delta); |
| 187 | int k = 0; |
| 188 | for (int j = 0; j < kGlobalSize; ++j) { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 189 | if (j == i) { |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 190 | EXPECT_EQ(x_plus_delta[j], x[j]); |
| 191 | } else { |
| 192 | EXPECT_EQ(x_plus_delta[j], x[j] + delta[k++]); |
| 193 | } |
| 194 | } |
| 195 | |
| 196 | double jacobian[kGlobalSize * kLocalSize]; |
| 197 | parameterization.ComputeJacobian(x, jacobian); |
| 198 | int delta_cursor = 0; |
| 199 | int jacobian_cursor = 0; |
| 200 | for (int j = 0; j < kGlobalSize; ++j) { |
| 201 | if (j != i) { |
| 202 | for (int k = 0; k < kLocalSize; ++k, jacobian_cursor++) { |
| 203 | EXPECT_EQ(jacobian[jacobian_cursor], delta_cursor == k ? 1.0 : 0.0); |
| 204 | } |
| 205 | ++delta_cursor; |
| 206 | } else { |
| 207 | for (int k = 0; k < kLocalSize; ++k, jacobian_cursor++) { |
| 208 | EXPECT_EQ(jacobian[jacobian_cursor], 0.0); |
| 209 | } |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | Matrix global_matrix = Matrix::Ones(10, kGlobalSize); |
| 214 | for (int row = 0; row < kGlobalSize; ++row) { |
| 215 | for (int col = 0; col < kGlobalSize; ++col) { |
| 216 | global_matrix(row, col) = col; |
| 217 | } |
| 218 | } |
| 219 | |
| 220 | Matrix local_matrix = Matrix::Zero(10, kLocalSize); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 221 | parameterization.MultiplyByJacobian( |
| 222 | x, 10, global_matrix.data(), local_matrix.data()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 223 | Matrix expected_local_matrix = |
| 224 | global_matrix * MatrixRef(jacobian, kGlobalSize, kLocalSize); |
| 225 | EXPECT_EQ((local_matrix - expected_local_matrix).norm(), 0.0); |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | // Functor needed to implement automatically differentiated Plus for |
| 230 | // quaternions. |
| 231 | struct QuaternionPlus { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 232 | template <typename T> |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 233 | bool operator()(const T* x, const T* delta, T* x_plus_delta) const { |
| 234 | const T squared_norm_delta = |
| 235 | delta[0] * delta[0] + delta[1] * delta[1] + delta[2] * delta[2]; |
| 236 | |
| 237 | T q_delta[4]; |
| 238 | if (squared_norm_delta > T(0.0)) { |
| 239 | T norm_delta = sqrt(squared_norm_delta); |
| 240 | const T sin_delta_by_delta = sin(norm_delta) / norm_delta; |
| 241 | q_delta[0] = cos(norm_delta); |
| 242 | q_delta[1] = sin_delta_by_delta * delta[0]; |
| 243 | q_delta[2] = sin_delta_by_delta * delta[1]; |
| 244 | q_delta[3] = sin_delta_by_delta * delta[2]; |
| 245 | } else { |
| 246 | // We do not just use q_delta = [1,0,0,0] here because that is a |
| 247 | // constant and when used for automatic differentiation will |
| 248 | // lead to a zero derivative. Instead we take a first order |
| 249 | // approximation and evaluate it at zero. |
| 250 | q_delta[0] = T(1.0); |
| 251 | q_delta[1] = delta[0]; |
| 252 | q_delta[2] = delta[1]; |
| 253 | q_delta[3] = delta[2]; |
| 254 | } |
| 255 | |
| 256 | QuaternionProduct(q_delta, x, x_plus_delta); |
| 257 | return true; |
| 258 | } |
| 259 | }; |
| 260 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 261 | template <typename Parameterization, typename Plus> |
| 262 | void QuaternionParameterizationTestHelper(const double* x, |
| 263 | const double* delta, |
| 264 | const double* x_plus_delta_ref) { |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 265 | const int kGlobalSize = 4; |
| 266 | const int kLocalSize = 3; |
| 267 | |
| 268 | const double kTolerance = 1e-14; |
| 269 | |
| 270 | double x_plus_delta[kGlobalSize] = {0.0, 0.0, 0.0, 0.0}; |
| 271 | Parameterization parameterization; |
| 272 | parameterization.Plus(x, delta, x_plus_delta); |
| 273 | for (int i = 0; i < kGlobalSize; ++i) { |
| 274 | EXPECT_NEAR(x_plus_delta[i], x_plus_delta[i], kTolerance); |
| 275 | } |
| 276 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 277 | const double x_plus_delta_norm = sqrt( |
| 278 | x_plus_delta[0] * x_plus_delta[0] + x_plus_delta[1] * x_plus_delta[1] + |
| 279 | x_plus_delta[2] * x_plus_delta[2] + x_plus_delta[3] * x_plus_delta[3]); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 280 | |
| 281 | EXPECT_NEAR(x_plus_delta_norm, 1.0, kTolerance); |
| 282 | |
| 283 | double jacobian_ref[12]; |
| 284 | double zero_delta[kLocalSize] = {0.0, 0.0, 0.0}; |
| 285 | const double* parameters[2] = {x, zero_delta}; |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 286 | double* jacobian_array[2] = {NULL, jacobian_ref}; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 287 | |
| 288 | // Autodiff jacobian at delta_x = 0. |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 289 | internal::AutoDifferentiate<kGlobalSize, |
| 290 | StaticParameterDims<kGlobalSize, kLocalSize>>( |
| 291 | Plus(), parameters, kGlobalSize, x_plus_delta, jacobian_array); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 292 | |
| 293 | double jacobian[12]; |
| 294 | parameterization.ComputeJacobian(x, jacobian); |
| 295 | for (int i = 0; i < 12; ++i) { |
| 296 | EXPECT_TRUE(IsFinite(jacobian[i])); |
| 297 | EXPECT_NEAR(jacobian[i], jacobian_ref[i], kTolerance) |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 298 | << "Jacobian mismatch: i = " << i << "\n Expected \n" |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 299 | << ConstMatrixRef(jacobian_ref, kGlobalSize, kLocalSize) |
| 300 | << "\n Actual \n" |
| 301 | << ConstMatrixRef(jacobian, kGlobalSize, kLocalSize); |
| 302 | } |
| 303 | |
| 304 | Matrix global_matrix = Matrix::Random(10, kGlobalSize); |
| 305 | Matrix local_matrix = Matrix::Zero(10, kLocalSize); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 306 | parameterization.MultiplyByJacobian( |
| 307 | x, 10, global_matrix.data(), local_matrix.data()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 308 | Matrix expected_local_matrix = |
| 309 | global_matrix * MatrixRef(jacobian, kGlobalSize, kLocalSize); |
| 310 | EXPECT_NEAR((local_matrix - expected_local_matrix).norm(), |
| 311 | 0.0, |
| 312 | 10.0 * std::numeric_limits<double>::epsilon()); |
| 313 | } |
| 314 | |
| 315 | template <int N> |
| 316 | void Normalize(double* x) { |
| 317 | VectorRef(x, N).normalize(); |
| 318 | } |
| 319 | |
| 320 | TEST(QuaternionParameterization, ZeroTest) { |
| 321 | double x[4] = {0.5, 0.5, 0.5, 0.5}; |
| 322 | double delta[3] = {0.0, 0.0, 0.0}; |
| 323 | double q_delta[4] = {1.0, 0.0, 0.0, 0.0}; |
| 324 | double x_plus_delta[4] = {0.0, 0.0, 0.0, 0.0}; |
| 325 | QuaternionProduct(q_delta, x, x_plus_delta); |
| 326 | QuaternionParameterizationTestHelper<QuaternionParameterization, |
| 327 | QuaternionPlus>(x, delta, x_plus_delta); |
| 328 | } |
| 329 | |
| 330 | TEST(QuaternionParameterization, NearZeroTest) { |
| 331 | double x[4] = {0.52, 0.25, 0.15, 0.45}; |
| 332 | Normalize<4>(x); |
| 333 | |
| 334 | double delta[3] = {0.24, 0.15, 0.10}; |
| 335 | for (int i = 0; i < 3; ++i) { |
| 336 | delta[i] = delta[i] * 1e-14; |
| 337 | } |
| 338 | |
| 339 | double q_delta[4]; |
| 340 | q_delta[0] = 1.0; |
| 341 | q_delta[1] = delta[0]; |
| 342 | q_delta[2] = delta[1]; |
| 343 | q_delta[3] = delta[2]; |
| 344 | |
| 345 | double x_plus_delta[4] = {0.0, 0.0, 0.0, 0.0}; |
| 346 | QuaternionProduct(q_delta, x, x_plus_delta); |
| 347 | QuaternionParameterizationTestHelper<QuaternionParameterization, |
| 348 | QuaternionPlus>(x, delta, x_plus_delta); |
| 349 | } |
| 350 | |
| 351 | TEST(QuaternionParameterization, AwayFromZeroTest) { |
| 352 | double x[4] = {0.52, 0.25, 0.15, 0.45}; |
| 353 | Normalize<4>(x); |
| 354 | |
| 355 | double delta[3] = {0.24, 0.15, 0.10}; |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 356 | const double delta_norm = |
| 357 | sqrt(delta[0] * delta[0] + delta[1] * delta[1] + delta[2] * delta[2]); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 358 | double q_delta[4]; |
| 359 | q_delta[0] = cos(delta_norm); |
| 360 | q_delta[1] = sin(delta_norm) / delta_norm * delta[0]; |
| 361 | q_delta[2] = sin(delta_norm) / delta_norm * delta[1]; |
| 362 | q_delta[3] = sin(delta_norm) / delta_norm * delta[2]; |
| 363 | |
| 364 | double x_plus_delta[4] = {0.0, 0.0, 0.0, 0.0}; |
| 365 | QuaternionProduct(q_delta, x, x_plus_delta); |
| 366 | QuaternionParameterizationTestHelper<QuaternionParameterization, |
| 367 | QuaternionPlus>(x, delta, x_plus_delta); |
| 368 | } |
| 369 | |
| 370 | // Functor needed to implement automatically differentiated Plus for |
| 371 | // Eigen's quaternion. |
| 372 | struct EigenQuaternionPlus { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 373 | template <typename T> |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 374 | bool operator()(const T* x, const T* delta, T* x_plus_delta) const { |
| 375 | const T norm_delta = |
| 376 | sqrt(delta[0] * delta[0] + delta[1] * delta[1] + delta[2] * delta[2]); |
| 377 | |
| 378 | Eigen::Quaternion<T> q_delta; |
| 379 | if (norm_delta > T(0.0)) { |
| 380 | const T sin_delta_by_delta = sin(norm_delta) / norm_delta; |
| 381 | q_delta.coeffs() << sin_delta_by_delta * delta[0], |
| 382 | sin_delta_by_delta * delta[1], sin_delta_by_delta * delta[2], |
| 383 | cos(norm_delta); |
| 384 | } else { |
| 385 | // We do not just use q_delta = [0,0,0,1] here because that is a |
| 386 | // constant and when used for automatic differentiation will |
| 387 | // lead to a zero derivative. Instead we take a first order |
| 388 | // approximation and evaluate it at zero. |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 389 | q_delta.coeffs() << delta[0], delta[1], delta[2], T(1.0); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 390 | } |
| 391 | |
| 392 | Eigen::Map<Eigen::Quaternion<T>> x_plus_delta_ref(x_plus_delta); |
| 393 | Eigen::Map<const Eigen::Quaternion<T>> x_ref(x); |
| 394 | x_plus_delta_ref = q_delta * x_ref; |
| 395 | return true; |
| 396 | } |
| 397 | }; |
| 398 | |
| 399 | TEST(EigenQuaternionParameterization, ZeroTest) { |
| 400 | Eigen::Quaterniond x(0.5, 0.5, 0.5, 0.5); |
| 401 | double delta[3] = {0.0, 0.0, 0.0}; |
| 402 | Eigen::Quaterniond q_delta(1.0, 0.0, 0.0, 0.0); |
| 403 | Eigen::Quaterniond x_plus_delta = q_delta * x; |
| 404 | QuaternionParameterizationTestHelper<EigenQuaternionParameterization, |
| 405 | EigenQuaternionPlus>( |
| 406 | x.coeffs().data(), delta, x_plus_delta.coeffs().data()); |
| 407 | } |
| 408 | |
| 409 | TEST(EigenQuaternionParameterization, NearZeroTest) { |
| 410 | Eigen::Quaterniond x(0.52, 0.25, 0.15, 0.45); |
| 411 | x.normalize(); |
| 412 | |
| 413 | double delta[3] = {0.24, 0.15, 0.10}; |
| 414 | for (int i = 0; i < 3; ++i) { |
| 415 | delta[i] = delta[i] * 1e-14; |
| 416 | } |
| 417 | |
| 418 | // Note: w is first in the constructor. |
| 419 | Eigen::Quaterniond q_delta(1.0, delta[0], delta[1], delta[2]); |
| 420 | |
| 421 | Eigen::Quaterniond x_plus_delta = q_delta * x; |
| 422 | QuaternionParameterizationTestHelper<EigenQuaternionParameterization, |
| 423 | EigenQuaternionPlus>( |
| 424 | x.coeffs().data(), delta, x_plus_delta.coeffs().data()); |
| 425 | } |
| 426 | |
| 427 | TEST(EigenQuaternionParameterization, AwayFromZeroTest) { |
| 428 | Eigen::Quaterniond x(0.52, 0.25, 0.15, 0.45); |
| 429 | x.normalize(); |
| 430 | |
| 431 | double delta[3] = {0.24, 0.15, 0.10}; |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 432 | const double delta_norm = |
| 433 | sqrt(delta[0] * delta[0] + delta[1] * delta[1] + delta[2] * delta[2]); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 434 | |
| 435 | // Note: w is first in the constructor. |
| 436 | Eigen::Quaterniond q_delta(cos(delta_norm), |
| 437 | sin(delta_norm) / delta_norm * delta[0], |
| 438 | sin(delta_norm) / delta_norm * delta[1], |
| 439 | sin(delta_norm) / delta_norm * delta[2]); |
| 440 | |
| 441 | Eigen::Quaterniond x_plus_delta = q_delta * x; |
| 442 | QuaternionParameterizationTestHelper<EigenQuaternionParameterization, |
| 443 | EigenQuaternionPlus>( |
| 444 | x.coeffs().data(), delta, x_plus_delta.coeffs().data()); |
| 445 | } |
| 446 | |
| 447 | // Functor needed to implement automatically differentiated Plus for |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 448 | // homogeneous vectors. |
| 449 | template <int Dim> |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 450 | struct HomogeneousVectorParameterizationPlus { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 451 | template <typename Scalar> |
| 452 | bool operator()(const Scalar* p_x, |
| 453 | const Scalar* p_delta, |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 454 | Scalar* p_x_plus_delta) const { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 455 | Eigen::Map<const Eigen::Matrix<Scalar, Dim, 1>> x(p_x); |
| 456 | Eigen::Map<const Eigen::Matrix<Scalar, Dim - 1, 1>> delta(p_delta); |
| 457 | Eigen::Map<Eigen::Matrix<Scalar, Dim, 1>> x_plus_delta(p_x_plus_delta); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 458 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 459 | const Scalar squared_norm_delta = delta.squaredNorm(); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 460 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 461 | Eigen::Matrix<Scalar, Dim, 1> y; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 462 | Scalar one_half(0.5); |
| 463 | if (squared_norm_delta > Scalar(0.0)) { |
| 464 | Scalar norm_delta = sqrt(squared_norm_delta); |
| 465 | Scalar norm_delta_div_2 = 0.5 * norm_delta; |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 466 | const Scalar sin_delta_by_delta = |
| 467 | sin(norm_delta_div_2) / norm_delta_div_2; |
| 468 | y.template head<Dim - 1>() = sin_delta_by_delta * one_half * delta; |
| 469 | y[Dim - 1] = cos(norm_delta_div_2); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 470 | |
| 471 | } else { |
| 472 | // We do not just use y = [0,0,0,1] here because that is a |
| 473 | // constant and when used for automatic differentiation will |
| 474 | // lead to a zero derivative. Instead we take a first order |
| 475 | // approximation and evaluate it at zero. |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 476 | y.template head<Dim - 1>() = delta * one_half; |
| 477 | y[Dim - 1] = Scalar(1.0); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 478 | } |
| 479 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 480 | Eigen::Matrix<Scalar, Dim, 1> v; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 481 | Scalar beta; |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 482 | |
| 483 | // NOTE: The explicit template arguments are needed here because |
| 484 | // ComputeHouseholderVector is templated and some versions of MSVC |
| 485 | // have trouble deducing the type of v automatically. |
| 486 | internal::ComputeHouseholderVector< |
| 487 | Eigen::Map<const Eigen::Matrix<Scalar, Dim, 1>>, |
| 488 | Scalar, |
| 489 | Dim>(x, &v, &beta); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 490 | |
| 491 | x_plus_delta = x.norm() * (y - v * (beta * v.dot(y))); |
| 492 | |
| 493 | return true; |
| 494 | } |
| 495 | }; |
| 496 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 497 | static void HomogeneousVectorParameterizationHelper(const double* x, |
| 498 | const double* delta) { |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 499 | const double kTolerance = 1e-14; |
| 500 | |
| 501 | HomogeneousVectorParameterization homogeneous_vector_parameterization(4); |
| 502 | |
| 503 | // Ensure the update maintains the norm. |
| 504 | double x_plus_delta[4] = {0.0, 0.0, 0.0, 0.0}; |
| 505 | homogeneous_vector_parameterization.Plus(x, delta, x_plus_delta); |
| 506 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 507 | const double x_plus_delta_norm = sqrt( |
| 508 | x_plus_delta[0] * x_plus_delta[0] + x_plus_delta[1] * x_plus_delta[1] + |
| 509 | x_plus_delta[2] * x_plus_delta[2] + x_plus_delta[3] * x_plus_delta[3]); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 510 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 511 | const double x_norm = |
| 512 | sqrt(x[0] * x[0] + x[1] * x[1] + x[2] * x[2] + x[3] * x[3]); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 513 | |
| 514 | EXPECT_NEAR(x_plus_delta_norm, x_norm, kTolerance); |
| 515 | |
| 516 | // Autodiff jacobian at delta_x = 0. |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 517 | AutoDiffLocalParameterization<HomogeneousVectorParameterizationPlus<4>, 4, 3> |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 518 | autodiff_jacobian; |
| 519 | |
| 520 | double jacobian_autodiff[12]; |
| 521 | double jacobian_analytic[12]; |
| 522 | |
| 523 | homogeneous_vector_parameterization.ComputeJacobian(x, jacobian_analytic); |
| 524 | autodiff_jacobian.ComputeJacobian(x, jacobian_autodiff); |
| 525 | |
| 526 | for (int i = 0; i < 12; ++i) { |
| 527 | EXPECT_TRUE(ceres::IsFinite(jacobian_analytic[i])); |
| 528 | EXPECT_NEAR(jacobian_analytic[i], jacobian_autodiff[i], kTolerance) |
| 529 | << "Jacobian mismatch: i = " << i << ", " << jacobian_analytic[i] << " " |
| 530 | << jacobian_autodiff[i]; |
| 531 | } |
| 532 | } |
| 533 | |
| 534 | TEST(HomogeneousVectorParameterization, ZeroTest) { |
| 535 | double x[4] = {0.0, 0.0, 0.0, 1.0}; |
| 536 | Normalize<4>(x); |
| 537 | double delta[3] = {0.0, 0.0, 0.0}; |
| 538 | |
| 539 | HomogeneousVectorParameterizationHelper(x, delta); |
| 540 | } |
| 541 | |
| 542 | TEST(HomogeneousVectorParameterization, NearZeroTest1) { |
| 543 | double x[4] = {1e-5, 1e-5, 1e-5, 1.0}; |
| 544 | Normalize<4>(x); |
| 545 | double delta[3] = {0.0, 1.0, 0.0}; |
| 546 | |
| 547 | HomogeneousVectorParameterizationHelper(x, delta); |
| 548 | } |
| 549 | |
| 550 | TEST(HomogeneousVectorParameterization, NearZeroTest2) { |
| 551 | double x[4] = {0.001, 0.0, 0.0, 0.0}; |
| 552 | double delta[3] = {0.0, 1.0, 0.0}; |
| 553 | |
| 554 | HomogeneousVectorParameterizationHelper(x, delta); |
| 555 | } |
| 556 | |
| 557 | TEST(HomogeneousVectorParameterization, AwayFromZeroTest1) { |
| 558 | double x[4] = {0.52, 0.25, 0.15, 0.45}; |
| 559 | Normalize<4>(x); |
| 560 | double delta[3] = {0.0, 1.0, -0.5}; |
| 561 | |
| 562 | HomogeneousVectorParameterizationHelper(x, delta); |
| 563 | } |
| 564 | |
| 565 | TEST(HomogeneousVectorParameterization, AwayFromZeroTest2) { |
| 566 | double x[4] = {0.87, -0.25, -0.34, 0.45}; |
| 567 | Normalize<4>(x); |
| 568 | double delta[3] = {0.0, 0.0, -0.5}; |
| 569 | |
| 570 | HomogeneousVectorParameterizationHelper(x, delta); |
| 571 | } |
| 572 | |
| 573 | TEST(HomogeneousVectorParameterization, AwayFromZeroTest3) { |
| 574 | double x[4] = {0.0, 0.0, 0.0, 2.0}; |
| 575 | double delta[3] = {0.0, 0.0, 0}; |
| 576 | |
| 577 | HomogeneousVectorParameterizationHelper(x, delta); |
| 578 | } |
| 579 | |
| 580 | TEST(HomogeneousVectorParameterization, AwayFromZeroTest4) { |
| 581 | double x[4] = {0.2, -1.0, 0.0, 2.0}; |
| 582 | double delta[3] = {1.4, 0.0, -0.5}; |
| 583 | |
| 584 | HomogeneousVectorParameterizationHelper(x, delta); |
| 585 | } |
| 586 | |
| 587 | TEST(HomogeneousVectorParameterization, AwayFromZeroTest5) { |
| 588 | double x[4] = {2.0, 0.0, 0.0, 0.0}; |
| 589 | double delta[3] = {1.4, 0.0, -0.5}; |
| 590 | |
| 591 | HomogeneousVectorParameterizationHelper(x, delta); |
| 592 | } |
| 593 | |
| 594 | TEST(HomogeneousVectorParameterization, DeathTests) { |
| 595 | EXPECT_DEATH_IF_SUPPORTED(HomogeneousVectorParameterization x(1), "size"); |
| 596 | } |
| 597 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 598 | // Functor needed to implement automatically differentiated Plus for |
| 599 | // line parameterization. |
| 600 | template <int AmbientSpaceDim> |
| 601 | struct LineParameterizationPlus { |
| 602 | template <typename Scalar> |
| 603 | bool operator()(const Scalar* p_x, |
| 604 | const Scalar* p_delta, |
| 605 | Scalar* p_x_plus_delta) const { |
| 606 | static constexpr int kTangetSpaceDim = AmbientSpaceDim - 1; |
| 607 | Eigen::Map<const Eigen::Matrix<Scalar, AmbientSpaceDim, 1>> origin_point( |
| 608 | p_x); |
| 609 | Eigen::Map<const Eigen::Matrix<Scalar, AmbientSpaceDim, 1>> dir( |
| 610 | p_x + AmbientSpaceDim); |
| 611 | Eigen::Map<const Eigen::Matrix<Scalar, kTangetSpaceDim, 1>> |
| 612 | delta_origin_point(p_delta); |
| 613 | Eigen::Map<Eigen::Matrix<Scalar, AmbientSpaceDim, 1>> |
| 614 | origin_point_plus_delta(p_x_plus_delta); |
| 615 | |
| 616 | HomogeneousVectorParameterizationPlus<AmbientSpaceDim> dir_plus; |
| 617 | dir_plus(dir.data(), |
| 618 | p_delta + kTangetSpaceDim, |
| 619 | p_x_plus_delta + AmbientSpaceDim); |
| 620 | |
| 621 | Eigen::Matrix<Scalar, AmbientSpaceDim, 1> v; |
| 622 | Scalar beta; |
| 623 | |
| 624 | // NOTE: The explicit template arguments are needed here because |
| 625 | // ComputeHouseholderVector is templated and some versions of MSVC |
| 626 | // have trouble deducing the type of v automatically. |
| 627 | internal::ComputeHouseholderVector< |
| 628 | Eigen::Map<const Eigen::Matrix<Scalar, AmbientSpaceDim, 1>>, |
| 629 | Scalar, |
| 630 | AmbientSpaceDim>(dir, &v, &beta); |
| 631 | |
| 632 | Eigen::Matrix<Scalar, AmbientSpaceDim, 1> y; |
| 633 | y << 0.5 * delta_origin_point, Scalar(0.0); |
| 634 | origin_point_plus_delta = origin_point + y - v * (beta * v.dot(y)); |
| 635 | |
| 636 | return true; |
| 637 | } |
| 638 | }; |
| 639 | |
| 640 | template <int AmbientSpaceDim> |
| 641 | static void LineParameterizationHelper(const double* x_ptr, |
| 642 | const double* delta) { |
| 643 | const double kTolerance = 1e-14; |
| 644 | |
| 645 | static constexpr int ParameterDim = 2 * AmbientSpaceDim; |
| 646 | static constexpr int TangientParameterDim = 2 * (AmbientSpaceDim - 1); |
| 647 | |
| 648 | LineParameterization<AmbientSpaceDim> line_parameterization; |
| 649 | |
| 650 | using ParameterVector = Eigen::Matrix<double, ParameterDim, 1>; |
| 651 | ParameterVector x_plus_delta = ParameterVector::Zero(); |
| 652 | line_parameterization.Plus(x_ptr, delta, x_plus_delta.data()); |
| 653 | |
| 654 | // Ensure the update maintains the norm for the line direction. |
| 655 | Eigen::Map<const ParameterVector> x(x_ptr); |
| 656 | const double dir_plus_delta_norm = |
| 657 | x_plus_delta.template tail<AmbientSpaceDim>().norm(); |
| 658 | const double dir_norm = x.template tail<AmbientSpaceDim>().norm(); |
| 659 | EXPECT_NEAR(dir_plus_delta_norm, dir_norm, kTolerance); |
| 660 | |
| 661 | // Ensure the update of the origin point is perpendicular to the line |
| 662 | // direction. |
| 663 | const double dot_prod_val = x.template tail<AmbientSpaceDim>().dot( |
| 664 | x_plus_delta.template head<AmbientSpaceDim>() - |
| 665 | x.template head<AmbientSpaceDim>()); |
| 666 | EXPECT_NEAR(dot_prod_val, 0.0, kTolerance); |
| 667 | |
| 668 | // Autodiff jacobian at delta_x = 0. |
| 669 | AutoDiffLocalParameterization<LineParameterizationPlus<AmbientSpaceDim>, |
| 670 | ParameterDim, |
| 671 | TangientParameterDim> |
| 672 | autodiff_jacobian; |
| 673 | |
| 674 | using JacobianMatrix = Eigen:: |
| 675 | Matrix<double, ParameterDim, TangientParameterDim, Eigen::RowMajor>; |
| 676 | constexpr double kNaN = std::numeric_limits<double>::quiet_NaN(); |
| 677 | JacobianMatrix jacobian_autodiff = JacobianMatrix::Constant(kNaN); |
| 678 | JacobianMatrix jacobian_analytic = JacobianMatrix::Constant(kNaN); |
| 679 | |
| 680 | autodiff_jacobian.ComputeJacobian(x_ptr, jacobian_autodiff.data()); |
| 681 | line_parameterization.ComputeJacobian(x_ptr, jacobian_analytic.data()); |
| 682 | |
| 683 | EXPECT_FALSE(jacobian_autodiff.hasNaN()); |
| 684 | EXPECT_FALSE(jacobian_analytic.hasNaN()); |
| 685 | EXPECT_TRUE(jacobian_autodiff.isApprox(jacobian_analytic)) |
| 686 | << "auto diff:\n" |
| 687 | << jacobian_autodiff << "\n" |
| 688 | << "analytic diff:\n" |
| 689 | << jacobian_analytic; |
| 690 | } |
| 691 | |
| 692 | TEST(LineParameterization, ZeroTest3D) { |
| 693 | double x[6] = {0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 694 | double delta[4] = {0.0, 0.0, 0.0, 0.0}; |
| 695 | |
| 696 | LineParameterizationHelper<3>(x, delta); |
| 697 | } |
| 698 | |
| 699 | TEST(LineParameterization, ZeroTest4D) { |
| 700 | double x[8] = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 701 | double delta[6] = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0}; |
| 702 | |
| 703 | LineParameterizationHelper<4>(x, delta); |
| 704 | } |
| 705 | |
| 706 | TEST(LineParameterization, ZeroOriginPointTest3D) { |
| 707 | double x[6] = {0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 708 | double delta[4] = {0.0, 0.0, 1.0, 2.0}; |
| 709 | |
| 710 | LineParameterizationHelper<3>(x, delta); |
| 711 | } |
| 712 | |
| 713 | TEST(LineParameterization, ZeroOriginPointTest4D) { |
| 714 | double x[8] = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 715 | double delta[6] = {0.0, 0.0, 0.0, 1.0, 2.0, 3.0}; |
| 716 | |
| 717 | LineParameterizationHelper<4>(x, delta); |
| 718 | } |
| 719 | |
| 720 | TEST(LineParameterization, ZeroDirTest3D) { |
| 721 | double x[6] = {0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 722 | double delta[4] = {3.0, 2.0, 0.0, 0.0}; |
| 723 | |
| 724 | LineParameterizationHelper<3>(x, delta); |
| 725 | } |
| 726 | |
| 727 | TEST(LineParameterization, ZeroDirTest4D) { |
| 728 | double x[8] = {0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0}; |
| 729 | double delta[6] = {3.0, 2.0, 1.0, 0.0, 0.0, 0.0}; |
| 730 | |
| 731 | LineParameterizationHelper<4>(x, delta); |
| 732 | } |
| 733 | |
| 734 | TEST(LineParameterization, AwayFromZeroTest3D1) { |
| 735 | Eigen::Matrix<double, 6, 1> x; |
| 736 | x.head<3>() << 1.54, 2.32, 1.34; |
| 737 | x.tail<3>() << 0.52, 0.25, 0.15; |
| 738 | x.tail<3>().normalize(); |
| 739 | |
| 740 | double delta[4] = {4.0, 7.0, 1.0, -0.5}; |
| 741 | |
| 742 | LineParameterizationHelper<3>(x.data(), delta); |
| 743 | } |
| 744 | |
| 745 | TEST(LineParameterization, AwayFromZeroTest4D1) { |
| 746 | Eigen::Matrix<double, 8, 1> x; |
| 747 | x.head<4>() << 1.54, 2.32, 1.34, 3.23; |
| 748 | x.tail<4>() << 0.52, 0.25, 0.15, 0.45; |
| 749 | x.tail<4>().normalize(); |
| 750 | |
| 751 | double delta[6] = {4.0, 7.0, -3.0, 0.0, 1.0, -0.5}; |
| 752 | |
| 753 | LineParameterizationHelper<4>(x.data(), delta); |
| 754 | } |
| 755 | |
| 756 | TEST(LineParameterization, AwayFromZeroTest3D2) { |
| 757 | Eigen::Matrix<double, 6, 1> x; |
| 758 | x.head<3>() << 7.54, -2.81, 8.63; |
| 759 | x.tail<3>() << 2.52, 5.25, 4.15; |
| 760 | |
| 761 | double delta[4] = {4.0, 7.0, 1.0, -0.5}; |
| 762 | |
| 763 | LineParameterizationHelper<3>(x.data(), delta); |
| 764 | } |
| 765 | |
| 766 | TEST(LineParameterization, AwayFromZeroTest4D2) { |
| 767 | Eigen::Matrix<double, 8, 1> x; |
| 768 | x.head<4>() << 7.54, -2.81, 8.63, 6.93; |
| 769 | x.tail<4>() << 2.52, 5.25, 4.15, 1.45; |
| 770 | |
| 771 | double delta[6] = {4.0, 7.0, -3.0, 2.0, 1.0, -0.5}; |
| 772 | |
| 773 | LineParameterizationHelper<4>(x.data(), delta); |
| 774 | } |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 775 | |
| 776 | class ProductParameterizationTest : public ::testing::Test { |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 777 | protected: |
| 778 | void SetUp() final { |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 779 | const int global_size1 = 5; |
| 780 | std::vector<int> constant_parameters1; |
| 781 | constant_parameters1.push_back(2); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 782 | param1_.reset( |
| 783 | new SubsetParameterization(global_size1, constant_parameters1)); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 784 | |
| 785 | const int global_size2 = 3; |
| 786 | std::vector<int> constant_parameters2; |
| 787 | constant_parameters2.push_back(0); |
| 788 | constant_parameters2.push_back(1); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 789 | param2_.reset( |
| 790 | new SubsetParameterization(global_size2, constant_parameters2)); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 791 | |
| 792 | const int global_size3 = 4; |
| 793 | std::vector<int> constant_parameters3; |
| 794 | constant_parameters3.push_back(1); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 795 | param3_.reset( |
| 796 | new SubsetParameterization(global_size3, constant_parameters3)); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 797 | |
| 798 | const int global_size4 = 2; |
| 799 | std::vector<int> constant_parameters4; |
| 800 | constant_parameters4.push_back(1); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 801 | param4_.reset( |
| 802 | new SubsetParameterization(global_size4, constant_parameters4)); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 803 | } |
| 804 | |
| 805 | std::unique_ptr<LocalParameterization> param1_; |
| 806 | std::unique_ptr<LocalParameterization> param2_; |
| 807 | std::unique_ptr<LocalParameterization> param3_; |
| 808 | std::unique_ptr<LocalParameterization> param4_; |
| 809 | }; |
| 810 | |
| 811 | TEST_F(ProductParameterizationTest, LocalAndGlobalSize2) { |
| 812 | LocalParameterization* param1 = param1_.release(); |
| 813 | LocalParameterization* param2 = param2_.release(); |
| 814 | |
| 815 | ProductParameterization product_param(param1, param2); |
| 816 | EXPECT_EQ(product_param.LocalSize(), |
| 817 | param1->LocalSize() + param2->LocalSize()); |
| 818 | EXPECT_EQ(product_param.GlobalSize(), |
| 819 | param1->GlobalSize() + param2->GlobalSize()); |
| 820 | } |
| 821 | |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 822 | TEST_F(ProductParameterizationTest, LocalAndGlobalSize3) { |
| 823 | LocalParameterization* param1 = param1_.release(); |
| 824 | LocalParameterization* param2 = param2_.release(); |
| 825 | LocalParameterization* param3 = param3_.release(); |
| 826 | |
| 827 | ProductParameterization product_param(param1, param2, param3); |
| 828 | EXPECT_EQ(product_param.LocalSize(), |
| 829 | param1->LocalSize() + param2->LocalSize() + param3->LocalSize()); |
| 830 | EXPECT_EQ(product_param.GlobalSize(), |
| 831 | param1->GlobalSize() + param2->GlobalSize() + param3->GlobalSize()); |
| 832 | } |
| 833 | |
| 834 | TEST_F(ProductParameterizationTest, LocalAndGlobalSize4) { |
| 835 | LocalParameterization* param1 = param1_.release(); |
| 836 | LocalParameterization* param2 = param2_.release(); |
| 837 | LocalParameterization* param3 = param3_.release(); |
| 838 | LocalParameterization* param4 = param4_.release(); |
| 839 | |
| 840 | ProductParameterization product_param(param1, param2, param3, param4); |
| 841 | EXPECT_EQ(product_param.LocalSize(), |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 842 | param1->LocalSize() + param2->LocalSize() + param3->LocalSize() + |
| 843 | param4->LocalSize()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 844 | EXPECT_EQ(product_param.GlobalSize(), |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 845 | param1->GlobalSize() + param2->GlobalSize() + param3->GlobalSize() + |
| 846 | param4->GlobalSize()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 847 | } |
| 848 | |
| 849 | TEST_F(ProductParameterizationTest, Plus) { |
| 850 | LocalParameterization* param1 = param1_.release(); |
| 851 | LocalParameterization* param2 = param2_.release(); |
| 852 | LocalParameterization* param3 = param3_.release(); |
| 853 | LocalParameterization* param4 = param4_.release(); |
| 854 | |
| 855 | ProductParameterization product_param(param1, param2, param3, param4); |
| 856 | std::vector<double> x(product_param.GlobalSize(), 0.0); |
| 857 | std::vector<double> delta(product_param.LocalSize(), 0.0); |
| 858 | std::vector<double> x_plus_delta_expected(product_param.GlobalSize(), 0.0); |
| 859 | std::vector<double> x_plus_delta(product_param.GlobalSize(), 0.0); |
| 860 | |
| 861 | for (int i = 0; i < product_param.GlobalSize(); ++i) { |
| 862 | x[i] = RandNormal(); |
| 863 | } |
| 864 | |
| 865 | for (int i = 0; i < product_param.LocalSize(); ++i) { |
| 866 | delta[i] = RandNormal(); |
| 867 | } |
| 868 | |
| 869 | EXPECT_TRUE(product_param.Plus(&x[0], &delta[0], &x_plus_delta_expected[0])); |
| 870 | int x_cursor = 0; |
| 871 | int delta_cursor = 0; |
| 872 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 873 | EXPECT_TRUE(param1->Plus( |
| 874 | &x[x_cursor], &delta[delta_cursor], &x_plus_delta[x_cursor])); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 875 | x_cursor += param1->GlobalSize(); |
| 876 | delta_cursor += param1->LocalSize(); |
| 877 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 878 | EXPECT_TRUE(param2->Plus( |
| 879 | &x[x_cursor], &delta[delta_cursor], &x_plus_delta[x_cursor])); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 880 | x_cursor += param2->GlobalSize(); |
| 881 | delta_cursor += param2->LocalSize(); |
| 882 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 883 | EXPECT_TRUE(param3->Plus( |
| 884 | &x[x_cursor], &delta[delta_cursor], &x_plus_delta[x_cursor])); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 885 | x_cursor += param3->GlobalSize(); |
| 886 | delta_cursor += param3->LocalSize(); |
| 887 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 888 | EXPECT_TRUE(param4->Plus( |
| 889 | &x[x_cursor], &delta[delta_cursor], &x_plus_delta[x_cursor])); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 890 | x_cursor += param4->GlobalSize(); |
| 891 | delta_cursor += param4->LocalSize(); |
| 892 | |
| 893 | for (int i = 0; i < x.size(); ++i) { |
| 894 | EXPECT_EQ(x_plus_delta[i], x_plus_delta_expected[i]); |
| 895 | } |
| 896 | } |
| 897 | |
| 898 | TEST_F(ProductParameterizationTest, ComputeJacobian) { |
| 899 | LocalParameterization* param1 = param1_.release(); |
| 900 | LocalParameterization* param2 = param2_.release(); |
| 901 | LocalParameterization* param3 = param3_.release(); |
| 902 | LocalParameterization* param4 = param4_.release(); |
| 903 | |
| 904 | ProductParameterization product_param(param1, param2, param3, param4); |
| 905 | std::vector<double> x(product_param.GlobalSize(), 0.0); |
| 906 | |
| 907 | for (int i = 0; i < product_param.GlobalSize(); ++i) { |
| 908 | x[i] = RandNormal(); |
| 909 | } |
| 910 | |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 911 | Matrix jacobian = |
| 912 | Matrix::Random(product_param.GlobalSize(), product_param.LocalSize()); |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 913 | EXPECT_TRUE(product_param.ComputeJacobian(&x[0], jacobian.data())); |
| 914 | int x_cursor = 0; |
| 915 | int delta_cursor = 0; |
| 916 | |
| 917 | Matrix jacobian1(param1->GlobalSize(), param1->LocalSize()); |
| 918 | EXPECT_TRUE(param1->ComputeJacobian(&x[x_cursor], jacobian1.data())); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 919 | jacobian.block( |
| 920 | x_cursor, delta_cursor, param1->GlobalSize(), param1->LocalSize()) -= |
| 921 | jacobian1; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 922 | x_cursor += param1->GlobalSize(); |
| 923 | delta_cursor += param1->LocalSize(); |
| 924 | |
| 925 | Matrix jacobian2(param2->GlobalSize(), param2->LocalSize()); |
| 926 | EXPECT_TRUE(param2->ComputeJacobian(&x[x_cursor], jacobian2.data())); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 927 | jacobian.block( |
| 928 | x_cursor, delta_cursor, param2->GlobalSize(), param2->LocalSize()) -= |
| 929 | jacobian2; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 930 | x_cursor += param2->GlobalSize(); |
| 931 | delta_cursor += param2->LocalSize(); |
| 932 | |
| 933 | Matrix jacobian3(param3->GlobalSize(), param3->LocalSize()); |
| 934 | EXPECT_TRUE(param3->ComputeJacobian(&x[x_cursor], jacobian3.data())); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 935 | jacobian.block( |
| 936 | x_cursor, delta_cursor, param3->GlobalSize(), param3->LocalSize()) -= |
| 937 | jacobian3; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 938 | x_cursor += param3->GlobalSize(); |
| 939 | delta_cursor += param3->LocalSize(); |
| 940 | |
| 941 | Matrix jacobian4(param4->GlobalSize(), param4->LocalSize()); |
| 942 | EXPECT_TRUE(param4->ComputeJacobian(&x[x_cursor], jacobian4.data())); |
Austin Schuh | 1d1e6ea | 2020-12-23 21:56:30 -0800 | [diff] [blame^] | 943 | jacobian.block( |
| 944 | x_cursor, delta_cursor, param4->GlobalSize(), param4->LocalSize()) -= |
| 945 | jacobian4; |
Austin Schuh | 70cc955 | 2019-01-21 19:46:48 -0800 | [diff] [blame] | 946 | x_cursor += param4->GlobalSize(); |
| 947 | delta_cursor += param4->LocalSize(); |
| 948 | |
| 949 | EXPECT_NEAR(jacobian.norm(), 0.0, std::numeric_limits<double>::epsilon()); |
| 950 | } |
| 951 | |
| 952 | } // namespace internal |
| 953 | } // namespace ceres |