#include "pycolmap/helpers.h" #include #include #include // Some important ceres bindings for using crucial pycolmap features (e.g. // bundle adjustment) without pyceres. // // Branched from // https://github.com/cvg/pyceres/blob/main/_pyceres/core/types.h // https://github.com/cvg/pyceres/blob/main/_pyceres/core/solver.h void BindCeresTypes(py::module& m) { auto ownt = py::enum_(m, "Ownership", py::module_local()) .value("DO_NOT_TAKE_OWNERSHIP", ceres::Ownership::DO_NOT_TAKE_OWNERSHIP) .value("TAKE_OWNERSHIP", ceres::Ownership::TAKE_OWNERSHIP) .export_values(); AddStringToEnumConstructor(ownt); auto mt = py::enum_(m, "MinimizerType", py::module_local()) .value("LINE_SEARCH", ceres::MinimizerType::LINE_SEARCH) .value("TRUST_REGION", ceres::MinimizerType::TRUST_REGION); AddStringToEnumConstructor(mt); auto linesearcht = py::enum_(m, "LineSearchType", py::module_local()) .value("ARMIJO", ceres::LineSearchType::ARMIJO) .value("WOLFE", ceres::LineSearchType::WOLFE); AddStringToEnumConstructor(linesearcht); auto lsdt = py::enum_( m, "LineSearchDirectionType", py::module_local()) .value("BFGS", ceres::LineSearchDirectionType::BFGS) .value("LBFGS", ceres::LineSearchDirectionType::LBFGS) .value("NONLINEAR_CONJUGATE_GRADIENT", ceres::LineSearchDirectionType::NONLINEAR_CONJUGATE_GRADIENT) .value("STEEPEST_DESCENT", ceres::LineSearchDirectionType::STEEPEST_DESCENT); AddStringToEnumConstructor(lsdt); auto lsit = py::enum_( m, "LineSearchInterpolationType", py::module_local()) .value("BISECTION", ceres::LineSearchInterpolationType::BISECTION) .value("CUBIC", ceres::LineSearchInterpolationType::CUBIC) .value("QUADRATIC", ceres::LineSearchInterpolationType::QUADRATIC); AddStringToEnumConstructor(lsit); auto ncgt = py::enum_( m, "NonlinearConjugateGradientType", py::module_local()) .value("FLETCHER_REEVES", ceres::NonlinearConjugateGradientType::FLETCHER_REEVES) .value("HESTENES_STIEFEL", ceres::NonlinearConjugateGradientType::HESTENES_STIEFEL) .value("POLAK_RIBIERE", ceres::NonlinearConjugateGradientType::POLAK_RIBIERE); AddStringToEnumConstructor(ncgt); auto linsolt = py::enum_( m, "LinearSolverType", py::module_local()) .value("DENSE_NORMAL_CHOLESKY", ceres::LinearSolverType::DENSE_NORMAL_CHOLESKY) .value("DENSE_QR", ceres::LinearSolverType::DENSE_QR) .value("SPARSE_NORMAL_CHOLESKY", ceres::LinearSolverType::SPARSE_NORMAL_CHOLESKY) .value("DENSE_SCHUR", ceres::LinearSolverType::DENSE_SCHUR) .value("SPARSE_SCHUR", ceres::LinearSolverType::SPARSE_SCHUR) .value("ITERATIVE_SCHUR", ceres::LinearSolverType::ITERATIVE_SCHUR) .value("CGNR", ceres::LinearSolverType::CGNR); AddStringToEnumConstructor(linsolt); auto dogt = py::enum_(m, "DoglegType", py::module_local()) .value("TRADITIONAL_DOGLEG", ceres::DoglegType::TRADITIONAL_DOGLEG) .value("SUBSPACE_DOGLEG", ceres::DoglegType::SUBSPACE_DOGLEG); AddStringToEnumConstructor(dogt); auto trst = py::enum_( m, "TrustRegionStrategyType", py::module_local()) .value("LEVENBERG_MARQUARDT", ceres::TrustRegionStrategyType::LEVENBERG_MARQUARDT) .value("DOGLEG", ceres::TrustRegionStrategyType::DOGLEG); AddStringToEnumConstructor(trst); auto prt = py::enum_( m, "PreconditionerType", py::module_local()) .value("IDENTITY", ceres::PreconditionerType::IDENTITY) .value("JACOBI", ceres::PreconditionerType::JACOBI) .value("SCHUR_JACOBI", ceres::PreconditionerType::SCHUR_JACOBI) .value("CLUSTER_JACOBI", ceres::PreconditionerType::CLUSTER_JACOBI) .value("CLUSTER_TRIDIAGONAL", ceres::PreconditionerType::CLUSTER_TRIDIAGONAL); AddStringToEnumConstructor(prt); auto vct = py::enum_( m, "VisibilityClusteringType", py::module_local()) .value("CANONICAL_VIEWS", ceres::VisibilityClusteringType::CANONICAL_VIEWS) .value("SINGLE_LINKAGE", ceres::VisibilityClusteringType::SINGLE_LINKAGE); AddStringToEnumConstructor(vct); auto dlalt = py::enum_( m, "DenseLinearAlgebraLibraryType", py::module_local()) .value("EIGEN", ceres::DenseLinearAlgebraLibraryType::EIGEN) .value("LAPACK", ceres::DenseLinearAlgebraLibraryType::LAPACK) .value("CUDA", ceres::DenseLinearAlgebraLibraryType::CUDA); AddStringToEnumConstructor(dlalt); auto slalt = py::enum_( m, "SparseLinearAlgebraLibraryType", py::module_local()) .value("SUITE_SPARSE", ceres::SparseLinearAlgebraLibraryType::SUITE_SPARSE) .value("EIGEN_SPARSE", ceres::SparseLinearAlgebraLibraryType::EIGEN_SPARSE) .value("ACCELERATE_SPARSE", ceres::SparseLinearAlgebraLibraryType::ACCELERATE_SPARSE) .value("NO_SPARSE", ceres::SparseLinearAlgebraLibraryType::NO_SPARSE); AddStringToEnumConstructor(slalt); auto logt = py::enum_(m, "LoggingType", py::module_local()) .value("SILENT", ceres::LoggingType::SILENT) .value("PER_MINIMIZER_ITERATION", ceres::LoggingType::PER_MINIMIZER_ITERATION); AddStringToEnumConstructor(logt); auto cbrt = py::enum_( m, "CallbackReturnType", py::module_local()) .value("SOLVER_CONTINUE", ceres::CallbackReturnType::SOLVER_CONTINUE) .value("SOLVER_ABORT", ceres::CallbackReturnType::SOLVER_ABORT) .value("SOLVER_TERMINATE_SUCCESSFULLY", ceres::CallbackReturnType::SOLVER_TERMINATE_SUCCESSFULLY); AddStringToEnumConstructor(cbrt); auto dft = py::enum_(m, "DumpFormatType", py::module_local()) .value("CONSOLE", ceres::DumpFormatType::CONSOLE) .value("TEXTFILE", ceres::DumpFormatType::TEXTFILE); AddStringToEnumConstructor(dft); auto termt = py::enum_( m, "TerminationType", py::module_local()) .value("CONVERGENCE", ceres::TerminationType::CONVERGENCE) .value("NO_CONVERGENCE", ceres::TerminationType::NO_CONVERGENCE) .value("FAILURE", ceres::TerminationType::FAILURE) .value("USER_SUCCESS", ceres::TerminationType::USER_SUCCESS) .value("USER_FAILURE", ceres::TerminationType::USER_FAILURE); AddStringToEnumConstructor(termt); } void BindCeresSolver(py::module& m) { using Options = ceres::Solver::Options; py::classh PyOptions(m, "SolverOptions", py::module_local()); PyOptions.def(py::init<>()) .def(py::init()) .def("IsValid", &Options::IsValid) .def_readwrite("minimizer_type", &Options::minimizer_type) .def_readwrite("line_search_direction_type", &Options::line_search_direction_type) .def_readwrite("line_search_type", &Options::line_search_type) .def_readwrite("nonlinear_conjugate_gradient_type", &Options::nonlinear_conjugate_gradient_type) .def_readwrite("max_lbfgs_rank", &Options::max_lbfgs_rank) .def_readwrite("use_approximate_eigenvalue_bfgs_scaling", &Options::use_approximate_eigenvalue_bfgs_scaling) .def_readwrite("line_search_interpolation_type", &Options::line_search_interpolation_type) .def_readwrite("min_line_search_step_size", &Options::min_line_search_step_size) .def_readwrite("line_search_sufficient_function_decrease", &Options::line_search_sufficient_function_decrease) .def_readwrite("max_line_search_step_contraction", &Options::max_line_search_step_contraction) .def_readwrite("min_line_search_step_contraction", &Options::min_line_search_step_contraction) .def_readwrite("max_num_line_search_step_size_iterations", &Options::max_num_line_search_step_size_iterations) .def_readwrite("max_num_line_search_direction_restarts", &Options::max_num_line_search_direction_restarts) .def_readwrite("line_search_sufficient_curvature_decrease", &Options::line_search_sufficient_curvature_decrease) .def_readwrite("max_line_search_step_expansion", &Options::max_line_search_step_expansion) .def_readwrite("trust_region_strategy_type", &Options::trust_region_strategy_type) .def_readwrite("dogleg_type", &Options::dogleg_type) .def_readwrite("use_nonmonotonic_steps", &Options::use_nonmonotonic_steps) .def_readwrite("max_consecutive_nonmonotonic_steps", &Options::max_consecutive_nonmonotonic_steps) .def_readwrite("max_num_iterations", &Options::max_num_iterations) .def_readwrite("max_solver_time_in_seconds", &Options::max_solver_time_in_seconds) .def_property( "num_threads", [](const Options& self) { return self.num_threads; }, [](Options& self, int n_threads) { int effective_n_threads = colmap::GetEffectiveNumThreads(n_threads); self.num_threads = effective_n_threads; #if CERES_VERSION_MAJOR < 2 self.num_linear_solver_threads = effective_n_threads; #endif // CERES_VERSION_MAJOR }) .def_readwrite("initial_trust_region_radius", &Options::initial_trust_region_radius) .def_readwrite("max_trust_region_radius", &Options::max_trust_region_radius) .def_readwrite("min_trust_region_radius", &Options::min_trust_region_radius) .def_readwrite("min_relative_decrease", &Options::min_relative_decrease) .def_readwrite("min_lm_diagonal", &Options::min_lm_diagonal) .def_readwrite("max_lm_diagonal", &Options::max_lm_diagonal) .def_readwrite("max_num_consecutive_invalid_steps", &Options::max_num_consecutive_invalid_steps) .def_readwrite("function_tolerance", &Options::function_tolerance) .def_readwrite("gradient_tolerance", &Options::gradient_tolerance) .def_readwrite("parameter_tolerance", &Options::parameter_tolerance) .def_readwrite("linear_solver_type", &Options::linear_solver_type) .def_readwrite("preconditioner_type", &Options::preconditioner_type) .def_readwrite("visibility_clustering_type", &Options::visibility_clustering_type) .def_readwrite("dense_linear_algebra_library_type", &Options::dense_linear_algebra_library_type) .def_readwrite("sparse_linear_algebra_library_type", &Options::sparse_linear_algebra_library_type) .def_readwrite("use_explicit_schur_complement", &Options::use_explicit_schur_complement) .def_readwrite("dynamic_sparsity", &Options::dynamic_sparsity) .def_readwrite("use_inner_iterations", &Options::use_inner_iterations) .def_readwrite("inner_iteration_tolerance", &Options::inner_iteration_tolerance) .def_readwrite("min_linear_solver_iterations", &Options::min_linear_solver_iterations) .def_readwrite("max_linear_solver_iterations", &Options::max_linear_solver_iterations) .def_readwrite("eta", &Options::eta) .def_readwrite("jacobi_scaling", &Options::jacobi_scaling) .def_readwrite("logging_type", &Options::logging_type) .def_readwrite("minimizer_progress_to_stdout", &Options::minimizer_progress_to_stdout) .def_readwrite("trust_region_problem_dump_directory", &Options::trust_region_problem_dump_directory) .def_readwrite("trust_region_problem_dump_format_type", &Options::trust_region_problem_dump_format_type) .def_readwrite("check_gradients", &Options::check_gradients) .def_readwrite("gradient_check_relative_precision", &Options::gradient_check_relative_precision) .def_readwrite( "gradient_check_numeric_derivative_relative_step_size", &Options::gradient_check_numeric_derivative_relative_step_size) .def_readwrite("update_state_every_iteration", &Options::update_state_every_iteration); MakeDataclass(PyOptions); using Summary = ceres::Solver::Summary; py::classh PySummary(m, "SolverSummary", py::module_local()); PySummary.def(py::init<>()) .def(py::init()) .def("BriefReport", &Summary::BriefReport) .def("FullReport", &Summary::FullReport) .def("IsSolutionUsable", &Summary::IsSolutionUsable) .def_readwrite("minimizer_type", &Summary::minimizer_type) .def_readwrite("termination_type", &Summary::termination_type) .def_readwrite("message", &Summary::message) .def_readwrite("initial_cost", &Summary::initial_cost) .def_readwrite("final_cost", &Summary::final_cost) .def_readwrite("fixed_cost", &Summary::fixed_cost) .def_readwrite("num_successful_steps", &Summary::num_successful_steps) .def_readwrite("num_unsuccessful_steps", &Summary::num_unsuccessful_steps) .def_readwrite("num_inner_iteration_steps", &Summary::num_inner_iteration_steps) .def_readwrite("num_line_search_steps", &Summary::num_line_search_steps) .def_readwrite("preprocessor_time_in_seconds", &Summary::preprocessor_time_in_seconds) .def_readwrite("minimizer_time_in_seconds", &Summary::minimizer_time_in_seconds) .def_readwrite("postprocessor_time_in_seconds", &Summary::postprocessor_time_in_seconds) .def_readwrite("total_time_in_seconds", &Summary::total_time_in_seconds) .def_readwrite("linear_solver_time_in_seconds", &Summary::linear_solver_time_in_seconds) .def_readwrite("num_linear_solves", &Summary::num_linear_solves) .def_readwrite("residual_evaluation_time_in_seconds", &Summary::residual_evaluation_time_in_seconds) .def_readwrite("num_residual_evaluations", &Summary::num_residual_evaluations) .def_readwrite("jacobian_evaluation_time_in_seconds", &Summary::jacobian_evaluation_time_in_seconds) .def_readwrite("num_jacobian_evaluations", &Summary::num_jacobian_evaluations) .def_readwrite("inner_iteration_time_in_seconds", &Summary::inner_iteration_time_in_seconds) .def_readwrite("line_search_cost_evaluation_time_in_seconds", &Summary::line_search_cost_evaluation_time_in_seconds) .def_readwrite("line_search_gradient_evaluation_time_in_seconds", &Summary::line_search_gradient_evaluation_time_in_seconds) .def_readwrite( "line_search_polynomial_minimization_time_in_seconds", &Summary::line_search_polynomial_minimization_time_in_seconds) .def_readwrite("line_search_total_time_in_seconds", &Summary::line_search_total_time_in_seconds) .def_readwrite("num_parameter_blocks", &Summary::num_parameter_blocks) .def_readwrite("num_parameters", &Summary::num_parameters) .def_readwrite("num_effective_parameters", &Summary::num_effective_parameters) .def_readwrite("num_residual_blocks", &Summary::num_residual_blocks) .def_readwrite("num_residuals", &Summary::num_residuals) .def_readwrite("num_parameter_blocks_reduced", &Summary::num_parameter_blocks_reduced) .def_readwrite("num_parameters_reduced", &Summary::num_parameters_reduced) .def_readwrite("num_effective_parameters_reduced", &Summary::num_effective_parameters_reduced) .def_readwrite("num_residual_blocks_reduced", &Summary::num_residual_blocks_reduced) .def_readwrite("num_residuals_reduced", &Summary::num_residuals_reduced) .def_readwrite("is_constrained", &Summary::is_constrained) .def_readwrite("num_threads_given", &Summary::num_threads_given) .def_readwrite("num_threads_used", &Summary::num_threads_used) #if CERES_VERSION_MAJOR < 2 .def_readwrite("num_linear_solver_threads_given", &Summary::num_linear_solver_threads_given) .def_readwrite("num_linear_solver_threads_used", &Summary::num_linear_solver_threads_used) #endif .def_readwrite("linear_solver_type_given", &Summary::linear_solver_type_given) .def_readwrite("linear_solver_type_used", &Summary::linear_solver_type_used) .def_readwrite("schur_structure_given", &Summary::schur_structure_given) .def_readwrite("schur_structure_used", &Summary::schur_structure_used) .def_readwrite("inner_iterations_given", &Summary::inner_iterations_given) .def_readwrite("inner_iterations_used", &Summary::inner_iterations_used) .def_readwrite("preconditioner_type_given", &Summary::preconditioner_type_given) .def_readwrite("preconditioner_type_used", &Summary::preconditioner_type_used) .def_readwrite("visibility_clustering_type", &Summary::visibility_clustering_type) .def_readwrite("trust_region_strategy_type", &Summary::trust_region_strategy_type) .def_readwrite("dogleg_type", &Summary::dogleg_type) .def_readwrite("dense_linear_algebra_library_type", &Summary::dense_linear_algebra_library_type) .def_readwrite("sparse_linear_algebra_library_type", &Summary::sparse_linear_algebra_library_type) .def_readwrite("line_search_direction_type", &Summary::line_search_direction_type) .def_readwrite("line_search_type", &Summary::line_search_type) .def_readwrite("line_search_interpolation_type", &Summary::line_search_interpolation_type) .def_readwrite("nonlinear_conjugate_gradient_type", &Summary::nonlinear_conjugate_gradient_type) .def_readwrite("max_lbfgs_rank", &Summary::max_lbfgs_rank); MakeDataclass(PySummary); using IterSummary = ceres::IterationSummary; py::classh PyIterSummary( m, "IterationSummary", py::module_local()); PyIterSummary.def(py::init<>()) .def(py::init()) .def_readonly("iteration", &IterSummary::iteration) .def_readonly("step_is_valid", &IterSummary::step_is_valid) .def_readonly("step_is_nonmonotonic", &IterSummary::step_is_nonmonotonic) .def_readonly("step_is_successful", &IterSummary::step_is_successful) .def_readonly("cost", &IterSummary::cost) .def_readonly("cost_change", &IterSummary::cost_change) .def_readonly("gradient_max_norm", &IterSummary::gradient_max_norm) .def_readonly("gradient_norm", &IterSummary::gradient_norm) .def_readonly("step_norm", &IterSummary::step_norm) .def_readonly("relative_decrease", &IterSummary::relative_decrease) .def_readonly("trust_region_radius", &IterSummary::trust_region_radius) .def_readonly("eta", &IterSummary::eta) .def_readonly("step_size", &IterSummary::step_size) .def_readonly("line_search_function_evaluations", &IterSummary::line_search_function_evaluations) .def_readonly("line_search_gradient_evaluations", &IterSummary::line_search_gradient_evaluations) .def_readonly("line_search_iterations", &IterSummary::line_search_iterations) .def_readonly("linear_solver_iterations", &IterSummary::linear_solver_iterations) .def_readonly("iteration_time_in_seconds", &IterSummary::iteration_time_in_seconds) .def_readonly("step_solver_time_in_seconds", &IterSummary::step_solver_time_in_seconds) .def_readonly("cumulative_time_in_seconds", &IterSummary::cumulative_time_in_seconds); } void BindCeres(py::module& m_parent) { py::module_ m = m_parent.def_submodule("pyceres"); BindCeresTypes(m); BindCeresSolver(m); }