Re: [ceres-solver] zero iteration and gradient max norm equal zeros

[Sameer Agarwal]

Miao,

This means there is something wrong with your gradient computation, which is returning zeros, and ceres thinks it has reached a local minimum.

One thing which is curious about your code is, that your constructor says you have parameter blocks of size 3, but you seem to be only using the first two elements of each parameter block. Is this correct?

Sameer

On Tue, Apr 10, 2018 at 6:23 PM Miao Yan <miaoy...@gmail.com> wrote:

I am building pos graph and trying to use ceres to solve the optimization problem. But the result keeps zero iteration and the gradient max norm reach zero immediately. I looked up online and think its jacobian is zero at the initial but do not know why. I attached the summary as following:

Solver Summary (v 1.11.0-eigen-(3.2.4)-no_lapack-eigensparse-openmp)

                                     Original                  Reduced

Parameter blocks                       137857                   137665

Parameters                             413571                   412995

Effective parameters                   138241                   137665

Residual blocks                        789264                   788425

Residual                              1578528                  1576850

Minimizer                        TRUST_REGION

Sparse linear algebra library    EIGEN_SPARSE

Trust region strategy     LEVENBERG_MARQUARDT

                                        Given                     Used

Linear solver          SPARSE_NORMAL_CHOLESKY   SPARSE_NORMAL_CHOLESKY

Threads                                     1                        1

Linear solver threads                       8                        8

Cost:

Initial                         6.515367e+011

Final                           6.515367e+011

Change                          0.000000e+000

Minimizer iterations                        0

Successful steps                            0

Unsuccessful steps                          0

Time (in seconds):

Preprocessor                           1.0601

  Residual evaluation                  0.0000

  Jacobian evaluation                  9.3984

  Linear solver                        0.0000

Minimizer                              9.4583

Postprocessor                          0.0538

Total                                 10.5721

Termination:                      CONVERGENCE (Gradient tolerance reached. Gradient max norm: 0.000000e+000 <= 0.000000e+000)

here is the costfunction:

struct xyerror_trail{

        const double x_diff_;

        const double y_diff_;

        const double weight_;

        xyerror_trail(const double& x_diff, const double& y_diff, const double& weight) :

            x_diff_(x_diff),

            y_diff_(y_diff),

            weight_(weight)

        {

        }

        static ceres::CostFunction* Create(const double& x_diff, const double& y_diff, const double& weight) {

            return (new ceres::AutoDiffCostFunction<xyerror_trail, 2, 3, 3>(new xyerror_trail(x_diff, y_diff, weight)));

        }

        template <typename T>

        bool operator()(const T* const veh_translation, const T* const veh_translation_dst, T* residuals) const {

            // Map T* to Eigen Vector3 with correct Scalar type

            Eigen::Matrix<T, 3, 1> t = Eigen::Map<const Eigen::Matrix<T, 3, 1> >(veh_translation);

            Eigen::Matrix<T, 3, 1> tDst = Eigen::Map<const Eigen::Matrix<T, 3, 1> >(veh_translation_dst);

            // The error is the difference between the predicted and observed position.

            residuals[0] = T(weight_) * (t[0] - tDst[0] - T(double(x_diff_)));

            residuals[1] = T(weight_) * (t[1] - tDst[1] - T(double(y_diff_ )));

            std::cout << residuals[0] << " " << residuals[1] << std::endl;

            return true;

        }

    };

basically, I have two types of xyerror_trail to build. First it is the trail itself and second is the relationship between trails. Since the first is trail self, most residuals are zeros at begining. Second ones are not zeros but have fewer residuals. Most residuals come from the first type. I thought maybe because most residuals are zeros, i made them not zeros but still not work. Anyone can help me ? Thanks. 

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/33ecab1e-25cd-4635-a0d5-f5eedb249466%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.