| 1 | /* ---------------------------------------------------------------------------- |
| 2 | |
| 3 | * GTSAM Copyright 2010, Georgia Tech Research Corporation, |
| 4 | * Atlanta, Georgia 30332-0415 |
| 5 | * All Rights Reserved |
| 6 | * Authors: Frank Dellaert, et al. (see THANKS for the full author list) |
| 7 | |
| 8 | * See LICENSE for the license information |
| 9 | |
| 10 | * -------------------------------------------------------------------------- */ |
| 11 | |
| 12 | /** |
| 13 | * @file SFMExample_SmartFactorPCG.cpp |
| 14 | * @brief Version of SFMExample_SmartFactor that uses Preconditioned Conjugate Gradient |
| 15 | * @author Frank Dellaert |
| 16 | */ |
| 17 | |
| 18 | // For an explanation of these headers, see SFMExample_SmartFactor.cpp |
| 19 | #include "SFMdata.h" |
| 20 | #include <gtsam/slam/SmartProjectionPoseFactor.h> |
| 21 | |
| 22 | // These extra headers allow us a LM outer loop with PCG linear solver (inner loop) |
| 23 | #include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h> |
| 24 | #include <gtsam/linear/Preconditioner.h> |
| 25 | #include <gtsam/linear/PCGSolver.h> |
| 26 | |
| 27 | using namespace std; |
| 28 | using namespace gtsam; |
| 29 | |
| 30 | // Make the typename short so it looks much cleaner |
| 31 | typedef SmartProjectionPoseFactor<Cal3_S2> SmartFactor; |
| 32 | |
| 33 | // create a typedef to the camera type |
| 34 | typedef PinholePose<Cal3_S2> Camera; |
| 35 | |
| 36 | /* ************************************************************************* */ |
| 37 | int main(int argc, char* argv[]) { |
| 38 | // Define the camera calibration parameters |
| 39 | Cal3_S2::shared_ptr K(new Cal3_S2(50.0, 50.0, 0.0, 50.0, 50.0)); |
| 40 | |
| 41 | // Define the camera observation noise model |
| 42 | auto measurementNoise = |
| 43 | noiseModel::Isotropic::Sigma(dim: 2, sigma: 1.0); // one pixel in u and v |
| 44 | |
| 45 | // Create the set of ground-truth landmarks and poses |
| 46 | vector<Point3> points = createPoints(); |
| 47 | vector<Pose3> poses = createPoses(); |
| 48 | |
| 49 | // Create a factor graph |
| 50 | NonlinearFactorGraph graph; |
| 51 | |
| 52 | // Simulated measurements from each camera pose, adding them to the factor graph |
| 53 | for (size_t j = 0; j < points.size(); ++j) { |
| 54 | // every landmark represent a single landmark, we use shared pointer to init |
| 55 | // the factor, and then insert measurements. |
| 56 | SmartFactor::shared_ptr smartfactor(new SmartFactor(measurementNoise, K)); |
| 57 | |
| 58 | for (size_t i = 0; i < poses.size(); ++i) { |
| 59 | // generate the 2D measurement |
| 60 | Camera camera(poses[i], K); |
| 61 | Point2 measurement = camera.project(pw: points[j]); |
| 62 | |
| 63 | // call add() function to add measurement into a single factor |
| 64 | smartfactor->add(measured: measurement, key: i); |
| 65 | } |
| 66 | |
| 67 | // insert the smart factor in the graph |
| 68 | graph.push_back(factor: smartfactor); |
| 69 | } |
| 70 | |
| 71 | // Add a prior on pose x0. This indirectly specifies where the origin is. |
| 72 | // 30cm std on x,y,z 0.1 rad on roll,pitch,yaw |
| 73 | auto noise = noiseModel::Diagonal::Sigmas( |
| 74 | sigmas: (Vector(6) << Vector3::Constant(value: 0.1), Vector3::Constant(value: 0.3)).finished()); |
| 75 | graph.addPrior(key: 0, prior: poses[0], model: noise); |
| 76 | |
| 77 | // Fix the scale ambiguity by adding a prior |
| 78 | graph.addPrior(key: 1, prior: poses[0], model: noise); |
| 79 | |
| 80 | // Create the initial estimate to the solution |
| 81 | Values initialEstimate; |
| 82 | Pose3 delta(Rot3::Rodrigues(wx: -0.1, wy: 0.2, wz: 0.25), Point3(0.05, -0.10, 0.20)); |
| 83 | for (size_t i = 0; i < poses.size(); ++i) |
| 84 | initialEstimate.insert(j: i, val: poses[i].compose(g: delta)); |
| 85 | |
| 86 | // We will use LM in the outer optimization loop, but by specifying |
| 87 | // "Iterative" below We indicate that an iterative linear solver should be |
| 88 | // used. In addition, the *type* of the iterativeParams decides on the type of |
| 89 | // iterative solver, in this case the SPCG (subgraph PCG) |
| 90 | LevenbergMarquardtParams parameters; |
| 91 | parameters.linearSolverType = NonlinearOptimizerParams::Iterative; |
| 92 | parameters.absoluteErrorTol = 1e-10; |
| 93 | parameters.relativeErrorTol = 1e-10; |
| 94 | parameters.maxIterations = 500; |
| 95 | PCGSolverParameters::shared_ptr pcg = |
| 96 | std::make_shared<PCGSolverParameters>(); |
| 97 | pcg->preconditioner = std::make_shared<BlockJacobiPreconditionerParameters>(); |
| 98 | // Following is crucial: |
| 99 | pcg->epsilon_abs = 1e-10; |
| 100 | pcg->epsilon_rel = 1e-10; |
| 101 | parameters.iterativeParams = pcg; |
| 102 | |
| 103 | LevenbergMarquardtOptimizer optimizer(graph, initialEstimate, parameters); |
| 104 | Values result = optimizer.optimize(); |
| 105 | |
| 106 | // Display result as in SFMExample_SmartFactor.run |
| 107 | result.print(str: "Final results:\n" ); |
| 108 | Values landmark_result; |
| 109 | for (size_t j = 0; j < points.size(); ++j) { |
| 110 | auto smart = std::dynamic_pointer_cast<SmartFactor>(r: graph[j]); |
| 111 | if (smart) { |
| 112 | std::optional<Point3> point = smart->point(values: result); |
| 113 | if (point) // ignore if std::optional return nullptr |
| 114 | landmark_result.insert(j, val: *point); |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | landmark_result.print(str: "Landmark results:\n" ); |
| 119 | cout << "final error: " << graph.error(values: result) << endl; |
| 120 | cout << "number of iterations: " << optimizer.iterations() << endl; |
| 121 | |
| 122 | return 0; |
| 123 | } |
| 124 | /* ************************************************************************* */ |
| 125 | |
| 126 | |