1 .. _chapter-version-history:
12 #. ``LineSearchMinimizer`` and ``GradientProblemSolver`` are up to 2x
13 faster due to fewer function evaluations. (Sameer Agarwal)
14 #. ``SPARSE_NORMAL_CHOLESKY`` is significantly faster because Ceres
15 now computes the normal equations exploiting the static block
16 sparsity structure. (Cheng Wang & Sameer Agarwal)
17 #. Add compound with scalar operators for Jets. (Alex Stewart)
18 #. Enable support for AVX instructions for Jets. (Alex Stewart)
20 Backward Incompatible API Changes
21 ---------------------------------
22 The enum ``CovarianceAlgorithmType`` which controls the linear algebra
23 algorithm used to compute the covariance used to combine the choice of
24 the algorithm and the choice of the sparse linear algebra library into
25 the enum name. So we had ``SUITE_SPARSE_QR`` and
26 ``EIGEN_SPARSE_QR``. ``Covariance::Options`` now has a separate member
27 allowing the user to choose the sparse linear algebra library, just
28 like the solver and ``CovarianceAlgorithmType`` now takes values
29 ``DENSE_SVD`` and ``SPARSE_QR``. This is a forward looking change that
30 will allow us to develop more flexible covariance estimation
31 algorithms with multiple linear algebra backends.
33 Bug Fixes & Minor Changes
34 -------------------------
35 #. Fix ``InvertPSDMatrix`` as it was triggering an Eigen assert in
36 Debug mode. (Philipp Hubner)
37 #. Fix cmake error from CeresConfig.cmake when Ceres not found (Taylor
39 #. Completely refactored ``SparseNormalCholeskySolver``. (Sameer
41 #. Fixed time reporting in ``Summary::FullReport`` when
42 ``LineSearchMinimizer`` is used. (Sameer Agarwal)
43 #. Remove unused file: collections_port.cc. (Sameer Agarwal)
44 #. ``SPARSE_SCHUR`` + ``CX_SPARSE`` = Faster (Sameer Agarwal)
45 #. Refactored a number of linear solver tests to be more thorough and
46 informative. (Sameer Agarwal)
47 #. Pass user-specified search hints as HINTS not PATHS. (Alex Stewart)
48 #. Prefer Eigen installs over exported build directories. (Alex
50 #. Add OpenMP flags when compiling for C if enabled. (Alex Stewart)
51 #. Add a missing ``CERES_EXPORT`` to GradientChecker (Sameer Agarwal)
52 #. Use target_compile_features() to specify C++11 requirement if
53 available. (Alex Stewart)
54 #. Update docs: .netrc --> .gitcookies (Keir Mierle)
55 #. Fix implicit precission loss warning on 64-bit archs (Ricardo
57 #. Optionally use exported Eigen CMake configuration if
58 available. (Alex Stewart)
59 #. Use ``Ceres_[SOURCE/BINARY]_DIR`` not ``CMAKE_XXX_DIR`` to support
60 nesting. (Alex Stewart)
61 #. Update ``Problem::EvaluateOptions`` documentation. (Sameer Agarwal)
62 #. Add public headers to CMake target for IDEs. (Devin Lane)
63 #. Add an article on interfacing with automatic
64 differentiation. (Sameer Agarwal)
65 #. Add default Fedora/Debian locations for CXSparse to search
67 #. Add a test for ``LineSearchMinimizer`` (Sameer Agarwal)
68 #. Flatten the table of contents. (Sameer Agarwal)
69 #. Fix when ``LineSearchMinimizer`` adds the ``IterationSummary``` to
70 ``Solver::Summary`` (Sameer Agarwal)
71 #. Fix search path for miniglog headers when Ceres is exported. (Alex
73 #. Fix ambiguous reference to ``WARNING`` when using miniglog. (Alex
75 #. Fix Jet/Eigen compatibility for Eigen > 3.3 (Julien Pilet)
76 #. Add max severity option when ``MINIGLOG`` is enabled (Taylor
78 #. Improvements to Schur template specializations (Sameer Agarwal)
79 #. Added an article on derivatives (Sameer Agarwal)
80 #. Require Eigen >= 3.3 to define ScalarBinaryOpTraits in Jet. (Alex
82 #. A hacky fix for the Eigen::FullPivLU changes. (Sameer Agarwal)
83 #. Specify ``ScalarBinaryOpTraits`` for Jet types. (Chris Sweeney)
84 #. Remove spurious conversion from doubles to Jets. (Sameer Agarwal)
85 #. Fix an error in the tutorial code for ``NumericDiffCostFunction``
87 #. ``CERES_EXPORT`` fix to compile Ceres as DLL (Je Hyeong Hong)
88 #. Fix detection of deprecated Bessel function names on MSVC. (Alex
90 #. Ensure that partial evaluation of residuals triggers an error
92 #. Fix detection of CMake-built glog on Windows. (Alex Stewart)
93 #. Add additional search paths for glog & Eigen on Windows. (Alex
95 #. Various minor grammar and bug fixes to the documentation (Sameer
96 Agarwal, Alex Stewart, William Rucklidge)
104 #. Aligned ``Jet`` matrices for improved automatic differentiation
105 performance. (Andrew Hunter)
106 #. Auto-differentiable implementations of Bessel functions, ``floor``,
107 and ``ceil`` (Alessandro Gentilini & Michael Vitus)
108 #. New 2D and 3D SLAM examples. (Michael Vitus)
109 #. Added ``EigenQuaternionParameterization``. (Michael Vitus)
110 #. Added ``Problem::IsParameterBlockConstant`` (Thomas Schneider)
111 #. A complete refactoring of ``TrustRegionMinimizer``. (Sameer Agarwal)
112 #. Gradient checking cleanup and local parameterization bugfix (David
116 Backward Incompatible API Changes
117 ---------------------------------
118 #. ``Solver::Options::numeric_derivative_relative_step_size`` has been
120 ``Solver::Options::gradient_check_numeric_derivative_relative_step_size``. (Sameer
123 Bug Fixes & Minor Changes
124 -------------------------
125 #. Clear XXX_FOUND in Find<XXX>.cmake prior to searching. (Alex
127 #. Fix versioning in the documentation (Sameer Agarwal)
128 #. Fix missing gflags imported target definition in
129 CeresConfig.cmake. (Alex Stewart)
130 #. Make gflags a public dependency of Ceres if it and glog are
131 found. (Alex Stewart)
132 #. Add support for glog exported CMake target. (Alex Stewart)
133 #. Use ``google::GLOG_WARNING`` instead of ``WARNING`` in tests to
134 support MSVC. (Alex Stewart)
135 #. Update gtest and gmock to
136 ``a2b8a8e07628e5fd60644b6dd99c1b5e7d7f1f47`` (Sameer Agarwal)
137 #. Add MSVC-specific ``#define`` to expose math constants in
138 ``<cmath>``. (Alex Stewart)
139 #. Fix typo. indepdendent -> independent (Hung Lun)
140 #. Fix potential invalid reset of CMAKE_FIND_LIBRARY_PREFIXES on MSVC
142 #. Fix use of alignas(0) which is not ignored on GCC (Alex Stewart)
143 #. Use default alignment if alignof(std::max_align_t) < 16 with C++11
145 #. Introduce a common base class for DynamicAutoDiffCostFunction and
146 DynamicNumericDiffCostFunction. (Sameer Agarwal)
147 #. Fix an exact equality test causing breakage in
148 gradient_checker_test. (Sameer Agarwal)
149 #. Add GradientProblemSolver::Options::parameter_tolerance. (Sameer
151 #. Add missing T() wrappers for constants. (Rob Carroll)
152 #. Remove two checks from rotation.h (Sameer Agarwal)
153 #. Relax the tolerance in QuaternionParameterizationTestHelper. (Je
155 #. Occured -> Occurred. (Sameer Agarwal)
156 #. Fix a test error in autodiff_test.cc. (Je Hyeong Hong)
157 #. Fix documentation source for templated function in ``rotation.h``.
158 #. Add ``package.xml`` to enable Catkin builds. (Damon Kohler)
159 #. Relaxing Jacobian matching in Gradient Checker test. (David Gossow)
160 #. Allow SubsetParameterization to hold all parameters constant
162 #. Fix an Intel compiler error in covariance_impl.cc (Je Hyeong Hong)
163 #. Removing duplicate include directive. (David Gossow)
164 #. Remove two DCHECKs from CubicHermiteSpline. (Sameer Agarwal)
165 #. Fix some compiler warnings. (Richard Trieu)
166 #. Update ExpectArraysClose to use ExpectClose instead of
167 EXPECT_NEAR. (Phillip Hubner)
168 #. FindWithDefault returns by value rather than reference. (@aradval)
169 #. Fix compiler errors on some systems. (David Gossow)
170 #. Note that Problem::Evaluate cannot be called from an
171 IterationCallback. (Sameer Agarwal)
172 #. Use ProductParameterization in bundle_adjuster.cc (Sameer Agarwal)
173 #. Enable support for OpenMP in Clang if detected. (Alex Stewart)
174 #. Remove duplicate entry for the NIST example in the docs. (Michael
176 #. Add additional logging for analyzing orderings (Sameer Agarwal)
177 #. Add readme for the sampled_function example. (Michael Vitus)
178 #. Use _j[0,1,n]() Bessel functions on MSVC to avoid deprecation
179 errors. (Alex Stewart & Kichang Kim)
180 #. Fix: Copy minimizer option ``is_silent`` to
181 ``LineSearchDirection::Options`` (Nicolai Wojke)
182 #. Fix typos in ``users.rst`` (Sameer Agarwal)
183 #. Make some Jet comparisons exact. (Sameer Agarwal)
184 #. Add colmap to users.rst (Sameer Agarwal)
185 #. Fix step norm evaluation in LineSearchMinimizer (Sameer Agarwal)
186 #. Remove use of -Werror when compiling Ceres. (Alex Stewart)
187 #. Report Ceres compile options as components in find_package(). (Alex
189 #. Fix a spelling error in nnls_modeling.rst (Timer)
190 #. Only use collapse() directive with OpenMP 3.0 or higher. (Keir
192 #. Fix install path for CeresConfig.cmake to be architecture-aware.
193 #. Fix double conversion to degrees in rotation_test (Keir Mierle)
194 #. Make Jet string output more readable (Keir Mierle)
195 #. Fix rotation_test IsClose() and related tests (Keir Mierle)
196 #. Loosen an exact equality in local_parameterization_test (Sameer
198 #. make_docs: Pass the file encoding to open() (Niels Ole Salscheider)
199 #. Fix error message returned when using SUITE_SPARSE_QR in covariance
200 estimation on a ceres built without SuiteSparse support. (Simon
202 #. Fix CXX11 option to be available on MinGW & CygWin, but not
204 #. Fix missing early return() in xxx_not_found() dependency
205 macros. (Alex Stewart)
206 #. Initialize ``inner_iterations_were_useful_`` correctly. (Sameer
208 #. Add an implementation for GradientProblemSolver::Options::IsValid
210 #. Fix use of va_copy() if compiling with explicit C++ version <
211 C++11. (Alex Stewart)
212 #. Install CMake files to lib/cmake/Ceres (Niels Ole Salscheider)
213 #. Allow users to override the documentation install directory. (Niels
215 #. Add covariance matrix for a vector of parameters (Wannes Van Loock)
216 #. Saner tolerances & stricter LRE test. (Sameer Agarwal)
217 #. Fix a malformed sentence in the tutorial. (Sameer Agarwal)
218 #. Add logging for sparse Cholesky factorization using Eigen. (Sameer
220 #. Use std::adjacent_find instead of std::unique. (Sameer Agarwal)
221 #. Improve logging in CompressedRowJacobianWriter on crash. (Sameer
223 #. Fix free parameter block handling in covariance computation (Wannes
225 #. Report the number of line search steps in FullReport. (Sameer
227 #. Make CMake read Ceres version directly from
228 include/ceres/version.h. (Alex Stewart)
229 #. Lots of code style/lint changes. (William Rucklidge)
230 #. Fix covariance computation for constant blocks (Wannes Van Loock)
231 #. Add IOS_DEPLOYMENT_TARGET variable to iOS.cmake (Eduard Feicho)
232 #. Make miniglog threadsafe on non-windows system by using
233 localtime_r() instead of localtime() for time formatting (Simon
241 #. Adaptive numeric differentiation using Ridders' method. (Tal
243 #. Add ``CubicInterpolator`` and ``BiCubicInterpolator`` to allow
244 smooth interpolation of sampled functions and integration with
245 automatic differentiation.
246 #. Add method to return covariance in tangent space. (Michael Vitus &
248 #. Add Homogeneous vector parameterization. (Michael Vitus)
249 #. Add a ``ProductParameterization``, a local parameterization that
250 can be constructed as a cartesian product of other local
252 #. Add DynamicCostFunctionToFunctor. (David Gossow)
253 #. Optionally export Ceres build directory into local CMake package
255 #. Faster ``SPARSE_NORMAL_CHOLESKY`` in the presence of dynamic
258 Bug Fixes & Minor Changes
259 -------------------------
260 #. Remove use of link-time optimisation (LTO) for all compilers due to
261 portability issues with gtest / type_info::operator== & Eigen with
262 Clang on OS X vs GCC 4.9+ on Linux requiring contradictory 'fixes'.
263 #. Use link-time optimisation (LTO) only when compiling Ceres itself,
264 not tests or examples, to bypass gtest / type_info::operator==
266 #. Use old minimum iOS version flags on Xcode < 7.0.
267 #. Add gtest-specific flags when building/using as a shared library.
268 #. Clean up iOS.cmake to use xcrun/xcodebuild & libtool.
269 #. Import the latest version of ``googletest``.
270 #. Refactored ``system_test`` into ``bundle_adjustment_test`` and
271 ``system_test``, where each test case is its own test.
272 #. Fix invalid memory access bug in
273 ``CompressedRowSparseMatrix::AppendRows`` when it was called with a
275 #. Build position independent code when compiling Ceres statically
277 #. Fix a bug in DetectStructure (Johannes Schonberger).
278 #. Reduce memory footprint of SubsetParameterization (Johannes
280 #. Fix for reorder program unit test when built without suitesparse
282 #. Fix a bug in the Schur eliminator (Werner Trobin).
283 #. Fix a bug in the reordering code (Bernhard Zeisl).
284 #. Add missing CERES_EXPORT to ComposedLoss (Simon Rutishauser).
285 #. Add the option to use numeric differentiation to ``nist`` and
286 ``more_garbow_hillstrom``.
287 #. Fix EIGENSPARSE option help s/t it displays in CMake ncurses GUI.
288 #. Fix SparseNormalCholeskySolver with dynamic sparsity (Richie
290 #. Remove legacy dependency detection macros.
291 #. Fix failed if() condition expansion if gflags is not found.
292 #. Update all CMake to lowercase function name style.
293 #. Update minimum iOS version to 7.0 for shared_ptr/unordered_map.
294 #. Fix bug in gflags' <= 2.1.2 exported CMake configuration.
295 #. Remove the spec file needed for generating RPMs.
296 #. Fix a typo in small_blas.h (Werber Trobin).
297 #. Cleanup FindGflags & use installed gflags CMake config if present.
298 #. Add default glog install location on Windows to search paths
300 #. Add default Eigen install location on Windows to search paths
302 #. Fix explanation of config.h generation in bare config.h.
303 #. Fix unused parameter compiler warnings in numeric_diff.h.
304 #. Increase tolerance for a test in polynomial_test (Taylor Braun
306 #. Fix addition of Gerrit commit hook when Ceres is a git submodule
308 #. Fix missing EIGEN_VERSION expansion typo.
309 #. Fix links to SuiteSparse & CXSparse (Henrique Mendonça).
310 #. Ensure Eigen is at least 3.1.0 for Eigen/SparseCore.
311 #. Add option to use C++11 (not TR1) shared_ptr & unordered_map
313 #. Fix an incorrect usage message in bundle_adjuster.cc
314 #. Gracefully disable docs if Sphinx is not found.
315 #. Explicitly use (new) default OS X rpath policy if present.
316 #. Add support of EIGEN_SPARSE type in
317 IsSparseLinearAlgebraLibraryTypeAvailable function (Pierre Moulon).
318 #. Allow the LossFunction contained in a LossFunctionWrapper to be
319 NULL. This is consistent with how NULL LossFunctions are treated
320 everywhere else. (Simon Rutishauser).
321 #. Improve numeric differentation near zero.
322 #. Refactored DynamicNumericDiffCostFunction to use NumericDiff (Tal
324 #. Remove use of :caption tag in Sphinx.
325 #. Add a small test to make sure GradientProblemSolver works correctly
327 #. Add simple unit tests for GradientProblem (Petter Strandmark).
328 #. Make the robust curve fitting example robust.
329 #. Homogenize convergence operators in docs and code (Johannes
331 #. Add parameter_tolerance convergence to line search minimizer
332 (Johannes Schonberger).
333 #. Fix bug where pow(JetA,JetB) returned wrong result for JetA==0
335 #. Remove duplicate step norm computation (Johannes Schonberger).
336 #. Enhance usability when encountering Eigen version mismatches
338 #. Add PLY file logger before and after BA in order to ease visual
339 comparison (Pierre Moulon).
340 #. Fix CMake config file docs to include 2.8.x & 3.x styles.
341 #. Python3 fixes (Markus Moll).
342 #. Remove confusing code from DenseJacobianWriter (Michael Vitus).
343 #. Add documentation on CMake package installation process.
344 #. Revert a call to SolveUpperTriangularUsingCholesky.
345 #. Make CERES_EIGEN_VERSION macro independent of CMake.
346 #. Add versions of dependencies used to FullReport().
347 #. Ensure local config.h is used if Ceres is already installed.
348 #. Small messaging and comment updates in CMake
349 #. Handle possible presence of library prefixes in MSVC (Sylvain
351 #. Use -O2 not -O3 on MinGW to workaround issue with Eigen
353 #. Increase tolerance in small_blas test for Cygwin
355 #. Fix iOS cmake file for cmake 3.0 (Jack Feng)
356 #. Fix missing gflags shlwapi dependency on MinGW (s1m3mu3@gmail.com).
357 #. Add thread dependency & fix namespace detection on Windows for
358 gflags (arrigo.benedetti@gmail.com).
359 #. Rename macros in the public API to have a ``CERES_`` prefix.
360 #. Fix ``OrderedGroup::Reverse()`` when it is empty (Chris Sweeney).
361 #. Update the code to point to ceres-solver.org.
362 #. Update documentation to point to the GitHub issue tracker.
363 #. Disable ``LAPACK`` for iOS builds. (Greg Coombe)
364 #. Force use of single-thread in ``Problem::Evaluate()`` without
366 #. Less strict check for multithreading. (Chris Sweeney)
367 #. Update tolerances in small_blas_test.cc (Philipp Hubner)
368 #. Documentation corrections (Steve Hsu)
369 #. Fixed ``sampled_function.cc`` (Pablo Speciale)
370 #. Fix example code in the documentation. (Rodney Hoskinson)
371 #. Improve the error handling in Conjugate Gradients.
372 #. Improve preconditioner documentation.
373 #. Remove dead code from fpclassify.h.
374 #. Make Android.mk threads sensitive.
375 #. Changed the ``CURRENT_CONFIG_INSTALL_DIR`` to be a variable local
376 to Ceres. (Chris Sweeney)
377 #. Fix typo in the comments in ``Jet.h``. (Julius Ziegler)
378 #. Add the ASL at ETH Zurich, Theia & OpenPTrack to the list of users.
379 #. Fixed a typo in the documentation. (Richard Stebbing)
380 #. Fixed a boundary handling bug in the BiCubic interpolation
381 code. (Bernhard Zeisl)
382 #. Fixed a ``MSVC`` compilation bug in the cubic interpolation code
383 (Johannes Schönberger)
384 #. Add covariance related files to the Android build.
385 #. Update Ubuntu 14.04 installation instructions. (Filippo Basso)
386 #. Improved logging for linear solver failures.
387 #. Improved crash messages in ``Problem``.
388 #. Hide Homebrew related variables in CMake GUI.
389 #. Add SuiteSparse link dependency for
390 compressed_col_sparse_matrix_utils_test.
391 #. Autodetect Homebrew install prefix on OSX.
392 #. Lint changes from William Rucklidge and Jim Roseborough.
393 #. Remove ``using namespace std:`` from ``port.h``
394 #. Add note about glog not currently compiling against gflags 2.1.
395 #. Add explicit no sparse linear algebra library available option.
396 #. Improve some wording in the FAQ. (Vasily Vylkov)
397 #. Delete Incomplete LQ Factorization.
398 #. Add a pointer to MacPorts. (Markus Moll)
406 #. Ceres Solver can now be used to solve general unconstrained
407 optimization problems. See the documentation for
408 ``GradientProblem`` and ``GradientProblemSolver``.
409 #. ``Eigen`` can now be as a sparse linear algebra backend. This can
411 ``Solver::Options::sparse_linear_algebra_library_type`` to
412 ``EIGEN_SPARSE``. Performance should be comparable to
417 Because ``Eigen`` is a header only library, and some of the code
418 related to sparse Cholesky factorization is LGPL, building Ceres
419 with support for Eigen's sparse linear algebra is disabled by
420 default and should be enabled explicitly.
424 For good performance, use Eigen version 3.2.2 or later.
426 #. Added ``EIGEN_SPARSE_QR`` algorithm for covariance estimation using
427 ``Eigen``'s sparse QR factorization. (Michael Vitus)
428 #. Faster inner iterations when using multiple threads.
429 #. Faster ``ITERATIVE_SCHUR`` + ``SCHUR_JACOBI`` for small to medium
430 sized problems (see documentation for
431 ``Solver::Options::use_explicit_schur_complement``).
432 #. Faster automatic Schur ordering.
433 #. Reduced memory usage when solving problems with dynamic sparsity.
434 #. ``CostFunctionToFunctor`` now supports dynamic number of residuals.
435 #. A complete re-write of the problem preprocessing phase.
436 #. ``Solver::Summary::FullReport`` now reports the build configuration
438 #. When building on Android, the ``NDK`` version detection logic has
440 #. The ``CERES_VERSION`` macro has been improved and replaced with the
441 ``CERES_VERSION_STRING`` macro.
442 #. Added ``Solver::Options::IsValid`` which allows users to validate
443 their solver configuration before calling ``Solve``.
444 #. Added ``Problem::GetCostFunctionForResidualBlock`` and
445 ``Problem::GetLossFunctionForResidualBlock``.
446 #. Added Tukey's loss function. (Michael Vitus)
447 #. Added RotationMatrixToQuaternion
448 #. Compute & report timing information for line searches.
449 #. Autodetect gflags namespace.
450 #. Expanded ``more_garbow_hillstrom.cc``.
451 #. Added a pointer to Tal Ben-Nun's MSVC wrapper to the docs.
452 #. Added the ``<2,3,6>`` Schur template specialization. (Alessandro
455 Backward Incompatible API Changes
456 ---------------------------------
457 #. ``NumericDiffFunctor`` has been removed. It's API was broken, and
458 the implementation was an unnecessary layer of abstraction over
459 ``CostFunctionToFunctor``.
460 #. ``POLAK_RIBIRERE`` conjugate gradients direction type has been
461 renamed to ``POLAK_RIBIERE``.
462 #. ``Solver::Options::solver_log`` has been removed. If needed this
463 iteration callback can easily be implemented in user code.
464 #. The ``SPARSE_CHOLESKY`` algorithm for covariance estimation has
465 been removed. It is not rank revealing and numerically poorly
466 behaved. Sparse QR factorization is a much better way to do this.
467 #. The ``SPARSE_QR`` algorithm for covariance estimation has been
468 renamed to ``SUITE_SPARSE_QR`` to be consistent with
470 #. ``Solver::Summary::preconditioner_type`` has been replaced with
471 ``Solver::Summary::preconditioner_type_given`` and
472 ``Solver::Summary::preconditioner_type_used`` to be more consistent
473 with how information about the linear solver is communicated.
474 #. ``CERES_VERSION`` and ``CERES_ABI_VERSION`` macros were not
475 terribly useful. They have been replaced with
476 ``CERES_VERSION_MAJOR``, ``CERES_VERSION_MINOR`` ,
477 ``CERES_VERSION_REVISION`` and ``CERES_VERSION_ABI`` macros. In
478 particular the functionality of ``CERES_VERSION`` is provided by
479 ``CERES_VERSION_STRING`` macro.
483 #. Do not try the gradient step if TR step line search fails.
484 #. Fix missing include in libmv_bundle_adjuster on OSX.
485 #. Conditionally log evaluation failure warnings.
486 #. Runtime uses four digits after the decimal in Summary:FullReport.
487 #. Better options checking for TrustRegionMinimizer.
488 #. Fix RotationMatrixToAngleAxis when the angle of rotation is near
490 #. Sometimes gradient norm based convergence would miss a step with a
491 substantial solution quality improvement. (Rodney Hoskinson)
492 #. Ignore warnings from within Eigen/SparseQR (3.2.2).
493 #. Fix empty Cache HELPSTRING parsing error on OS X 10.10 Yosemite.
494 #. Fix a formatting error TrustRegionMinimizer logging.
495 #. Add an explicit include for local_parameterization.h (cooordz)
496 #. Fix a number of typos in the documentation (Martin Baeuml)
497 #. Made the logging in TrustRegionMinimizer consistent with
499 #. Fix some obsolete documentation in CostFunction::Evaluate.
500 #. Fix CG solver options for ITERATIVE_SCHUR, which did not copy
501 min_num_iterations (Johannes Schönberger)
502 #. Remove obsolete include of numeric_diff_functor.h. (Martin Baeuml)
503 #. Fix max. linear solver iterations in ConjugateGradientsSolver
504 (Johannes Schönberger)
505 #. Expand check for lack of a sparse linear algebra library. (Michael
506 Samples and Domink Reitzle)
507 #. Fix Eigen Row/ColMajor bug in NumericDiffCostFunction. (Dominik
509 #. Fix crash in Covariance if # threads > 1 requested without OpenMP.
510 #. Fixed Malformed regex. (Björn Piltz)
511 #. Fixed MSVC error C2124: divide or mod by zero. (Björn Piltz)
512 #. Add missing #include of <limits> for loss functions.
513 #. Make canned loss functions more robust.
514 #. Fix type of suppressed compiler warning for Eigen 3.2.0.
515 #. Suppress unused variable warning from Eigen 3.2.0.
516 #. Add "make install" to the install instructions.
517 #. Correct formula in documentation of
518 Solver::Options::function_tolerance. (Alessandro Gentilini)
519 #. Add release flags to iOS toolchain.
520 #. Fix a broken hyperlink in the documentation. (Henrique Mendonca)
521 #. Add fixes for multiple definitions of ERROR on Windows to docs.
522 #. Compile miniglog into Ceres if enabled on all platforms.
523 #. Add two missing files to Android.mk (Greg Coombe)
524 #. Fix Cmake error when using miniglog. (Greg Coombe)
525 #. Don't build miniglog unconditionally as a static library (Björn
527 #. Added a missing include. (Björn Piltz)
528 #. Conditionally disable SparseNormalCholesky.
529 #. Fix a memory leak in program_test.cc.
537 #. Bounds constraints: Support for upper and/or lower bounds on
538 parameters when using the trust region minimizer.
539 #. Dynamic Sparsity: Problems in which the sparsity structure of the
540 Jacobian changes over the course of the optimization can now be
541 solved much more efficiently. (Richard Stebbing)
542 #. Improved support for Microsoft Visual C++ including the ability to
543 build and ship DLLs. (Björn Piltz, Alex Stewart and Sergey
545 #. Support for building on iOS 6.0 or higher (Jack Feng).
546 #. Autogeneration of config.h that captures all the defines used to
547 build and use Ceres Solver.
548 #. Simpler and more informative solver termination type
549 reporting. (See below for more details)
550 #. New `website <http://www.ceres-solver.org>`_ based entirely on
552 #. ``AutoDiffLocalParameterization`` allows the use of automatic
553 differentiation for defining ``LocalParameterization`` objects
555 #. LBFGS is faster due to fewer memory copies.
556 #. Parameter blocks are not restricted to be less than 32k in size,
557 they can be up to 2G in size.
558 #. Faster ``SPARSE_NORMAL_CHOLESKY`` solver when using ``CX_SPARSE``
559 as the sparse linear algebra library.
560 #. Added ``Problem::IsParameterBlockPresent`` and
561 ``Problem::GetParameterization``.
562 #. Added the (2,4,9) and (2,4,8) template specializations.
563 #. An example demonstrating the use of
564 DynamicAutoDiffCostFunction. (Joydeep Biswas)
565 #. Homography estimation example from Blender demonstrating the use of
566 a custom ``IterationCallback``. (Sergey Sharybin)
567 #. Support user passing a custom CMAKE_MODULE_PATH (for BLAS /
570 Backward Incompatible API Changes
571 ---------------------------------
572 #. ``Solver::Options::linear_solver_ordering`` used to be a naked
573 pointer that Ceres took ownership of. This is error prone behaviour
574 which leads to problems when copying the ``Solver::Options`` struct
575 around. This has been replaced with a ``shared_ptr`` to handle
576 ownership correctly across copies.
578 #. The enum used for reporting the termination/convergence status of
579 the solver has been renamed from ``SolverTerminationType`` to
582 The enum values have also changed. ``FUNCTION_TOLERANCE``,
583 ``GRADIENT_TOLERANCE`` and ``PARAMETER_TOLERANCE`` have all been
584 replaced by ``CONVERGENCE``.
586 ``NUMERICAL_FAILURE`` has been replaed by ``FAILURE``.
588 ``USER_ABORT`` has been renamed to ``USER_FAILURE``.
590 Further ``Solver::Summary::error`` has been renamed to
591 ``Solver::Summary::message``. It contains a more detailed
592 explanation for why the solver terminated.
594 #. ``Solver::Options::gradient_tolerance`` used to be a relative
595 gradient tolerance. i.e., The solver converged when
597 .. math:: \|g(x)\|_\infty < \text{gradient_tolerance} *
600 where :math:`g(x)` is the gradient of the objective function at
601 :math:`x` and :math:`x_0` is the parmeter vector at the start of
604 This has changed to an absolute tolerance, i.e. the solver
607 .. math:: \|g(x)\|_\infty < \text{gradient_tolerance}
609 #. Ceres cannot be built without the line search minimizer
610 anymore. Thus the preprocessor define
611 ``CERES_NO_LINE_SEARCH_MINIMIZER`` has been removed.
615 #. Disabled warning C4251. (Björn Piltz)
616 #. Do not propagate 3d party libs through
617 `IMPORTED_LINK_INTERFACE_LIBRARIES_[DEBUG/RELEASE]` mechanism when
618 building shared libraries. (Björn Piltz)
619 #. Fixed errant verbose levels (Björn Piltz)
620 #. Variety of code cleanups, optimizations and bug fixes to the line
621 search minimizer code (Alex Stewart)
622 #. Fixed ``BlockSparseMatrix::Transpose`` when the matrix has row and
623 column blocks. (Richard Bowen)
624 #. Better error checking when ``Problem::RemoveResidualBlock`` is
625 called. (Alex Stewart)
626 #. Fixed a memory leak in ``SchurComplementSolver``.
627 #. Added ``epsilon()`` method to ``NumTraits<ceres::Jet<T, N>
629 #. Fixed a bug in `CompressedRowSparseMatrix::AppendRows`` and
631 #. Handle empty problems consistently.
632 #. Restore the state of the ``Problem`` after a call to
633 ``Problem::Evaluate``. (Stefan Leutenegger)
634 #. Better error checking and reporting for linear solvers.
635 #. Use explicit formula to solve quadratic polynomials instead of the
637 #. Fix constant parameter handling in inner iterations (Mikael
639 #. SuiteSparse errors do not cause a fatal crash anymore.
640 #. Fix ``corrector_test.cc``.
641 #. Relax the requirements on loss function derivatives.
642 #. Minor bugfix to logging.h (Scott Ettinger)
643 #. Updated ``gmock`` and ``gtest`` to the latest upstream version.
644 #. Fix build breakage on old versions of SuiteSparse.
645 #. Fixed build issues related to Clang / LLVM 3.4 (Johannes
647 #. METIS_FOUND is never set. Changed the commit to fit the setting of
648 the other #._FOUND definitions. (Andreas Franek)
649 #. Variety of bug fixes and cleanups to the ``CMake`` build system
651 #. Removed fictious shared library target from the NDK build.
652 #. Solver::Options now uses ``shared_ptr`` to handle ownership of
653 ``Solver::Options::linear_solver_ordering`` and
654 ``Solver::Options::inner_iteration_ordering``. As a consequence the
655 ``NDK`` build now depends on ``libc++`` from the ``LLVM`` project.
656 #. Variety of lint cleanups (William Rucklidge & Jim Roseborough)
657 #. Various internal cleanups including dead code removal.
665 #. Significant improved ``CMake`` files with better robustness,
666 dependency checking and GUI support. (Alex Stewart)
667 #. Added ``DynamicNumericDiffCostFunction`` for numerically
668 differentiated cost functions whose sizing is determined at run
670 #. ``NumericDiffCostFunction`` now supports a dynamic number of
671 residuals just like ``AutoDiffCostFunction``.
672 #. ``Problem`` exposes more of its structure in its API.
673 #. Faster automatic differentiation (Tim Langlois)
674 #. Added the commonly occuring ``2_d_d`` template specialization for
675 the Schur Eliminator.
676 #. Faster ``ITERATIVE_SCHUR`` solver using template specializations.
677 #. Faster ``SCHUR_JACOBI`` preconditioner construction.
678 #. Faster ``AngleAxisRotatePoint``.
679 #. Faster Jacobian evaluation when a loss function is used.
680 #. Added support for multiple clustering algorithms in visibility
681 based preconditioning, including a new fast single linkage
682 clustering algorithm.
686 #. Fix ordering of ParseCommandLineFlags() & InitGoogleTest() for
687 Windows. (Alex Stewart)
688 #. Remove DCHECK_GE checks from fixed_array.h.
689 #. Fix build on MSVC 2013 (Petter Strandmark)
690 #. Fixed ``AngleAxisToRotationMatrix`` near zero.
691 #. Move ``CERES_HASH_NAMESPACE`` macros to ``collections_port.h``.
692 #. Fix handling of unordered_map/unordered_set on OSX 10.9.0.
693 #. Explicitly link to libm for ``curve_fitting_c.c``. (Alex Stewart)
694 #. Minor type conversion fix to autodiff.h
695 #. Remove RuntimeNumericDiffCostFunction.
696 #. Fix operator= ambiguity on some versions of Clang. (Alex Stewart)
697 #. Various Lint cleanups (William Rucklidge & Jim Roseborough)
698 #. Modified installation folders for Windows. (Pablo Speciale)
699 #. Added librt to link libraries for SuiteSparse_config on
700 Linux. (Alex Stewart)
701 #. Check for presence of return-type-c-linkage option with
702 Clang. (Alex Stewart)
703 #. Fix Problem::RemoveParameterBlock after calling solve. (Simon
705 #. Fix a free/delete bug in covariance_impl.cc
706 #. Fix two build errors. (Dustin Lang)
707 #. Add RequireInitialization = 1 to NumTraits::Jet.
708 #. Update gmock/gtest to 1.7.0
709 #. Added IterationSummary::gradient_norm.
710 #. Reduced verbosity of the inner iteration minimizer.
711 #. Fixed a bug in TrustRegionMinimizer. (Michael Vitus)
712 #. Removed android/build_android.sh.
718 Backward Incompatible API Changes
719 ---------------------------------
721 #. ``Solver::Options::sparse_linear_algebra_library`` has been renamed
722 to ``Solver::Options::sparse_linear_algebra_library_type``.
726 #. Sparse and dense covariance estimation.
727 #. A new Wolfe line search. (Alex Stewart)
728 #. ``BFGS`` line search direction. (Alex Stewart)
730 #. Speeded up the use of loss functions > 17x.
731 #. Faster ``DENSE_QR``, ``DENSE_NORMAL_CHOLESKY`` and ``DENSE_SCHUR``
733 #. Support for multiple dense linear algebra backends. In particular
734 optimized ``BLAS`` and ``LAPACK`` implementations (e.g., Intel MKL,
735 ACML, OpenBLAS etc) can now be used to do the dense linear algebra
736 for ``DENSE_QR``, ``DENSE_NORMAL_CHOLESKY`` and ``DENSE_SCHUR``
737 #. Use of Inner iterations can now be adaptively stopped. Iteration
738 and runtime statistics for inner iterations are not reported in
739 ``Solver::Summary`` and ``Solver::Summary::FullReport``.
740 #. Improved inner iteration step acceptance criterion.
741 #. Add BlockRandomAccessCRSMatrix.
742 #. Speeded up automatic differentiation by 7\%.
743 #. Bundle adjustment example from libmv/Blender (Sergey Sharybin)
744 #. Shared library building is now controlled by CMake, rather than a
745 custom solution. Previously, Ceres had a custom option, but this is
746 now deprecated in favor of CMake's built in support for switching
747 between static and shared. Turn on BUILD_SHARED_LIBS to get shared
749 #. No more dependence on Protocol Buffers.
750 #. Incomplete LQ factorization.
751 #. Ability to write trust region problems to disk.
752 #. Add sinh, cosh, tanh and tan functions to automatic differentiation
753 (Johannes Schönberger)
754 #. Simplifications to the cmake build file.
755 #. ``miniglog`` can now be used as a replacement for ``google-glog``
756 on non Android platforms. (This is NOT recommended).
760 #. Fix ``ITERATIVE_SCHUR`` solver to work correctly when the schur
761 complement is of size zero. (Soohyun Bae)
762 #. Fix the ``spec`` file for generating ``RPM`` packages (Brian Pitts
763 and Taylor Braun-Jones).
764 #. Fix how ceres calls CAMD (Manas Jagadev)
765 #. Fix breakage on old versions of SuiteSparse. (Fisher Yu)
766 #. Fix warning C4373 in Visual Studio (Petter Strandmark)
767 #. Fix compilation error caused by missing suitesparse headers and
768 reorganize them to be more robust. (Sergey Sharybin)
769 #. Check GCC Version before adding -fast compiler option on
770 OSX. (Steven Lovegrove)
771 #. Add documentation for minimizer progress output.
772 #. Lint and other cleanups (William Rucklidge and James Roseborough)
773 #. Collections port fix for MSC 2008 (Sergey Sharybin)
774 #. Various corrections and cleanups in the documentation.
775 #. Change the path where CeresConfig.cmake is installed (Pablo
777 #. Minor errors in documentation (Pablo Speciale)
778 #. Updated depend.cmake to follow CMake IF convention. (Joydeep
780 #. Stablize the schur ordering algorithm.
781 #. Update license header in split.h.
782 #. Enabling -O4 (link-time optimization) only if compiler/linker
783 support it. (Alex Stewart)
784 #. Consistent glog path across files.
785 #. ceres-solver.spec: Use cleaner, more conventional Release string
787 #. Fix compile bug on RHEL6 due to missing header (Taylor Braun-Jones)
788 #. CMake file is less verbose.
789 #. Use the latest upstream version of google-test and gmock.
790 #. Rationalize some of the variable names in ``Solver::Options``.
791 #. Improve Summary::FullReport when line search is used.
792 #. Expose line search parameters in ``Solver::Options``.
793 #. Fix update of L-BFGS history buffers after they become full. (Alex
795 #. Fix configuration error on systems without SuiteSparse installed
797 #. Enforce the read call returns correct value in
798 ``curve_fitting_c.c`` (Arnaud Gelas)
799 #. Fix DynamicAutoDiffCostFunction (Richard Stebbing)
800 #. Fix Problem::RemoveParameterBlock documentation (Johannes
802 #. Fix a logging bug in parameter_block.h
803 #. Refactor the preconditioner class structure.
804 #. Fix an uninitialized variable warning when building with ``GCC``.
805 #. Fix a reallocation bug in
806 ``CreateJacobianBlockSparsityTranspose``. (Yuliy Schwartzburg)
807 #. Add a define for O_BINARY.
808 #. Fix miniglog-based Android NDK build; now works with NDK r9. (Scott
817 #. Major Performance improvements.
819 a. Schur type solvers (``SPARSE_SCHUR``, ``DENSE_SCHUR``,
820 ``ITERATIVE_SCHUR``) are significantly faster due to custom BLAS
821 routines and fewer heap allocations.
823 b. ``SPARSE_SCHUR`` when used with ``CX_SPARSE`` now uses a block
824 AMD for much improved factorization performance.
826 c. The jacobian matrix is pre-ordered so that
827 ``SPARSE_NORMAL_CHOLESKY`` and ``SPARSE_SCHUR`` do not have to
828 make copies inside ``CHOLMOD``.
830 d. Faster autodiff by replacing division by multplication by inverse.
832 e. When compiled without threads, the schur eliminator does not pay
833 the penalty for locking and unlocking mutexes.
835 #. Users can now use ``linear_solver_ordering`` to affect the
836 fill-reducing ordering used by ``SUITE_SPARSE`` for
837 ``SPARSE_NORMAL_CHOLESKY``.
838 #. ``Problem`` can now report the set of parameter blocks it knows about.
839 #. ``TrustRegionMinimizer`` uses the evaluator to compute the gradient
840 instead of a matrix vector multiply.
841 #. On ``Mac OS``, whole program optimization is enabled.
842 #. Users can now use automatic differentiation to define new
843 ``LocalParameterization`` objects. (Sergey Sharybin)
844 #. Enable larger tuple sizes for Visual Studio 2012. (Petter Strandmark)
850 #. Update the documentation for ``CostFunction``.
851 #. Fixed a typo in the documentation. (Pablo Speciale)
852 #. Fix a typo in suitesparse.cc.
853 #. Bugfix in ``NumericDiffCostFunction``. (Nicolas Brodu)
854 #. Death to BlockSparseMatrixBase.
855 #. Change Minimizer::Options::min_trust_region_radius to double.
856 #. Update to compile with stricter gcc checks. (Joydeep Biswas)
857 #. Do not modify cached CMAKE_CXX_FLAGS_RELEASE. (Sergey Sharybin)
858 #. ``<iterator>`` needed for back_insert_iterator. (Petter Strandmark)
859 #. Lint cleanup. (William Rucklidge)
860 #. Documentation corrections. (Pablo Speciale)
866 Backward Incompatible API Changes
867 ---------------------------------
868 #. Added ``Problem::Evaluate``. Now you can evaluate a problem or any
869 part of it without calling the solver.
871 In light of this the following settings have been deprecated and
872 removed from the API.
874 - ``Solver::Options::return_initial_residuals``
875 - ``Solver::Options::return_initial_gradient``
876 - ``Solver::Options::return_initial_jacobian``
877 - ``Solver::Options::return_final_residuals``
878 - ``Solver::Options::return_final_gradient``
879 - ``Solver::Options::return_final_jacobian``
881 Instead we recommend using something like this.
888 vector<double> initial_residuals;
889 problem.Evaluate(Problem::EvaluateOptions(),
892 NULL, /* No gradient */
893 NULL /* No jacobian */ );
895 Solver::Options options;
896 Solver::Summary summary;
897 Solver::Solve(options, &problem, &summary);
899 vector<double> final_residuals;
900 problem.Evaluate(Problem::EvaluateOptions(),
903 NULL, /* No gradient */
904 NULL /* No jacobian */ );
909 #. Problem now supports removal of ParameterBlocks and
910 ResidualBlocks. There is a space/time tradeoff in doing this which
912 ``Problem::Options::enable_fast_parameter_block_removal``.
914 #. Ceres now supports Line search based optimization algorithms in
915 addition to trust region algorithms. Currently there is support for
916 gradient descent, non-linear conjugate gradient and LBFGS search
918 #. Added ``Problem::Evaluate``. Now you can evaluate a problem or any
919 part of it without calling the solver. In light of this the
920 following settings have been deprecated and removed from the API.
922 - ``Solver::Options::return_initial_residuals``
923 - ``Solver::Options::return_initial_gradient``
924 - ``Solver::Options::return_initial_jacobian``
925 - ``Solver::Options::return_final_residuals``
926 - ``Solver::Options::return_final_gradient``
927 - ``Solver::Options::return_final_jacobian``
929 #. New, much improved HTML documentation using Sphinx.
930 #. Changed ``NumericDiffCostFunction`` to take functors like
931 ``AutoDiffCostFunction``.
932 #. Added support for mixing automatic, analytic and numeric
933 differentiation. This is done by adding ``CostFunctionToFunctor``
934 and ``NumericDiffFunctor`` objects to the API.
935 #. Sped up the robust loss function correction logic when residual is
937 #. Sped up ``DenseQRSolver`` by changing the way dense jacobians are
938 stored. This is a 200-500% improvement in linear solver performance
939 depending on the size of the problem.
940 #. ``DENSE_SCHUR`` now supports multi-threading.
941 #. Greatly expanded ``Summary::FullReport``:
943 - Report the ordering used by the ``LinearSolver``.
944 - Report the ordering used by the inner iterations.
945 - Execution timing breakdown into evaluations and linear solves.
946 - Effective size of the problem solved by the solver, which now
947 accounts for the size of the tangent space when using a
948 ``LocalParameterization``.
949 #. Ceres when run at the ``VLOG`` level 3 or higher will report
950 detailed timing information about its internals.
951 #. Remove extraneous initial and final residual evaluations. This
952 speeds up the solver a bit.
953 #. Automatic differenatiation with a dynamic number of parameter
954 blocks. (Based on an idea by Thad Hughes).
955 #. Sped up problem construction and destruction.
956 #. Added matrix adapters to ``rotation.h`` so that the rotation matrix
957 routines can work with row and column major matrices. (Markus Moll)
958 #. ``SCHUR_JACOBI`` can now be used without ``SuiteSparse``.
959 #. A ``.spec`` file for producing RPMs. (Taylor Braun-Jones)
960 #. ``CMake`` can now build the sphinx documentation (Pablo Speciale)
961 #. Add support for creating a CMake config file during build to make
962 embedding Ceres in other CMake-using projects easier. (Pablo
964 #. Better error reporting in ``Problem`` for missing parameter blocks.
965 #. A more flexible ``Android.mk`` and a more modular build. If binary
966 size and/or compile time is a concern, larger parts of the solver
967 can be disabled at compile time.
971 #. Compilation fixes for MSVC2010 (Sergey Sharybin)
972 #. Fixed "deprecated conversion from string constant to char*"
973 warnings. (Pablo Speciale)
974 #. Correctly propagate ifdefs when building without Schur eliminator
975 template specializations.
976 #. Correct handling of ``LIB_SUFFIX`` on Linux. (Yuliy Schwartzburg).
977 #. Code and signature cleanup in ``rotation.h``.
978 #. Make examples independent of internal code.
979 #. Disable unused member in ``gtest`` which results in build error on
980 OS X with latest Xcode. (Taylor Braun-Jones)
981 #. Pass the correct flags to the linker when using
982 ``pthreads``. (Taylor Braun-Jones)
983 #. Only use ``cmake28`` macro when building on RHEL6. (Taylor
985 #. Remove ``-Wno-return-type-c-linkage`` when compiling with
986 GCC. (Taylor Braun-Jones)
987 #. Fix ``No previous prototype`` warnings. (Sergey Sharybin)
988 #. MinGW build fixes. (Sergey Sharybin)
989 #. Lots of minor code and lint fixes. (William Rucklidge)
990 #. Fixed a bug in ``solver_impl.cc`` residual evaluation. (Markus
992 #. Fixed varidic evaluation bug in ``AutoDiff``.
993 #. Fixed ``SolverImpl`` tests.
994 #. Fixed a bug in ``DenseSparseMatrix::ToDenseMatrix()``.
995 #. Fixed an initialization bug in ``ProgramEvaluator``.
996 #. Fixes to Android.mk paths (Carlos Hernandez)
997 #. Modify ``nist.cc`` to compute accuracy based on ground truth
998 solution rather than the ground truth function value.
999 #. Fixed a memory leak in ``cxsparse.cc``. (Alexander Mordvintsev).
1000 #. Fixed the install directory for libraries by correctly handling
1001 ``LIB_SUFFIX``. (Taylor Braun-Jones)
1006 Backward Incompatible API Changes
1007 ---------------------------------
1008 The new ordering API breaks existing code. Here the common case fixes.
1014 options.linear_solver_type = ceres::DENSE_SCHUR
1015 options.ordering_type = ceres::SCHUR
1022 options.linear_solver_type = ceres::DENSE_SCHUR
1029 options.linear_solver_type = ceres::DENSE_SCHUR;
1030 options.ordering_type = ceres::USER;
1031 for (int i = 0; i < num_points; ++i) {
1032 options.ordering.push_back(my_points[i])
1034 for (int i = 0; i < num_cameras; ++i) {
1035 options.ordering.push_back(my_cameras[i])
1037 options.num_eliminate_blocks = num_points;
1044 options.linear_solver_type = ceres::DENSE_SCHUR;
1045 options.ordering = new ceres::ParameterBlockOrdering;
1046 for (int i = 0; i < num_points; ++i) {
1047 options.linear_solver_ordering->AddElementToGroup(my_points[i], 0);
1049 for (int i = 0; i < num_cameras; ++i) {
1050 options.linear_solver_ordering->AddElementToGroup(my_cameras[i], 1);
1056 #. A new richer, more expressive and consistent API for ordering
1058 #. A non-linear generalization of Ruhe & Wedin's Algorithm II. This
1059 allows the user to use variable projection on separable and
1060 non-separable non-linear least squares problems. With
1061 multithreading, this results in significant improvements to the
1062 convergence behavior of the solver at a small increase in run time.
1063 #. An image denoising example using fields of experts. (Petter
1065 #. Defines for Ceres version and ABI version.
1066 #. Higher precision timer code where available. (Petter Strandmark)
1067 #. Example Makefile for users of Ceres.
1068 #. IterationSummary now informs the user when the step is a
1070 #. Fewer memory allocations when using ``DenseQRSolver``.
1071 #. GradientChecker for testing CostFunctions (William Rucklidge)
1072 #. Add support for cost functions with 10 parameter blocks in
1073 ``Problem``. (Fisher)
1074 #. Add support for 10 parameter blocks in ``AutoDiffCostFunction``.
1080 #. static cast to force Eigen::Index to long conversion
1081 #. Change LOG(ERROR) to LOG(WARNING) in ``schur_complement_solver.cc``.
1082 #. Remove verbose logging from ``DenseQRSolve``.
1083 #. Fix the Android NDK build.
1084 #. Better handling of empty and constant Problems.
1085 #. Remove an internal header that was leaking into the public API.
1086 #. Memory leak in ``trust_region_minimizer.cc``
1087 #. Schur ordering was operating on the wrong object (Ricardo Martin)
1088 #. MSVC fixes (Petter Strandmark)
1089 #. Various fixes to ``nist.cc`` (Markus Moll)
1090 #. Fixed a jacobian scaling bug.
1091 #. Numerically robust computation of ``model_cost_change``.
1092 #. Signed comparison compiler warning fixes (Ricardo Martin)
1093 #. Various compiler warning fixes all over.
1094 #. Inclusion guard fixes (Petter Strandmark)
1095 #. Segfault in test code (Sergey Popov)
1096 #. Replaced ``EXPECT/ASSERT_DEATH`` with the more portable
1097 ``EXPECT_DEATH_IF_SUPPORTED`` macros.
1098 #. Fixed the camera projection model in Ceres' implementation of
1099 Snavely's camera model. (Ricardo Martin)
1107 #. Android Port (Scott Ettinger also contributed to the port)
1108 #. Windows port. (Changchang Wu and Pierre Moulon also contributed to the port)
1109 #. New subspace Dogleg Solver. (Markus Moll)
1110 #. Trust region algorithm now supports the option of non-monotonic steps.
1111 #. New loss functions ``ArcTanLossFunction``, ``TolerantLossFunction``
1112 and ``ComposedLossFunction``. (James Roseborough).
1113 #. New ``DENSE_NORMAL_CHOLESKY`` linear solver, which uses Eigen's
1114 LDLT factorization on the normal equations.
1115 #. Cached symbolic factorization when using ``CXSparse``.
1117 #. New example ``nist.cc`` and data from the NIST non-linear
1118 regression test suite. (Thanks to Douglas Bates for suggesting this.)
1119 #. The traditional Dogleg solver now uses an elliptical trust
1120 region (Markus Moll)
1121 #. Support for returning initial and final gradients & Jacobians.
1122 #. Gradient computation support in the evaluators, with an eye
1123 towards developing first order/gradient based solvers.
1124 #. A better way to compute ``Solver::Summary::fixed_cost``. (Markus Moll)
1125 #. ``CMake`` support for building documentation, separate examples,
1126 installing and uninstalling the library and Gerrit hooks (Arnaud
1128 #. ``SuiteSparse4`` support (Markus Moll)
1129 #. Support for building Ceres without ``TR1`` (This leads to
1130 slightly slower ``DENSE_SCHUR`` and ``SPARSE_SCHUR`` solvers).
1131 #. ``BALProblem`` can now write a problem back to disk.
1132 #. ``bundle_adjuster`` now allows the user to normalize and perturb the
1133 problem before solving.
1134 #. Solver progress logging to file.
1135 #. Added ``Program::ToString`` and ``ParameterBlock::ToString`` to
1136 help with debugging.
1137 #. Ability to build Ceres as a shared library (MacOS and Linux only),
1138 associated versioning and build release script changes.
1139 #. Portable floating point classification API.
1144 #. Fix how invalid step evaluations are handled.
1145 #. Change the slop handling around zero for model cost changes to use
1146 relative tolerances rather than absolute tolerances.
1147 #. Fix an inadvertant integer to bool conversion. (Petter Strandmark)
1148 #. Do not link to ``libgomp`` when building on
1149 windows. (Petter Strandmark)
1150 #. Include ``gflags.h`` in ``test_utils.cc``. (Petter
1152 #. Use standard random number generation routines. (Petter Strandmark)
1153 #. ``TrustRegionMinimizer`` does not implicitly negate the
1154 steps that it takes. (Markus Moll)
1155 #. Diagonal scaling allows for equal upper and lower bounds. (Markus Moll)
1156 #. TrustRegionStrategy does not misuse LinearSolver:Summary anymore.
1157 #. Fix Eigen3 Row/Column Major storage issue. (Lena Gieseke)
1158 #. QuaternionToAngleAxis now guarantees an angle in $[-\pi, \pi]$. (Guoxuan Zhang)
1159 #. Added a workaround for a compiler bug in the Android NDK to the
1161 #. The sparse linear algebra library is only logged in
1162 Summary::FullReport if it is used.
1163 #. Rename the macro ``CERES_DONT_HAVE_PROTOCOL_BUFFERS``
1164 to ``CERES_NO_PROTOCOL_BUFFERS`` for consistency.
1165 #. Fix how static structure detection for the Schur eliminator logs
1167 #. Correct example code in the documentation. (Petter Strandmark)
1168 #. Fix ``fpclassify.h`` to work with the Android NDK and STLport.
1169 #. Fix a memory leak in the ``levenber_marquardt_strategy_test.cc``
1170 #. Fix an early return bug in the Dogleg solver. (Markus Moll)
1171 #. Zero initialize Jets.
1172 #. Moved ``internal/ceres/mock_log.h`` to ``internal/ceres/gmock/mock-log.h``
1173 #. Unified file path handling in tests.
1174 #. ``data_fitting.cc`` includes ``gflags``
1175 #. Renamed Ceres' Mutex class and associated macros to avoid
1176 namespace conflicts.
1177 #. Close the BAL problem file after reading it (Markus Moll)
1178 #. Fix IsInfinite on Jets.
1179 #. Drop alignment requirements for Jets.
1180 #. Fixed Jet to integer comparison. (Keith Leung)
1181 #. Fix use of uninitialized arrays. (Sebastian Koch & Markus Moll)
1182 #. Conditionally compile gflag dependencies.(Casey Goodlett)
1183 #. Add ``data_fitting.cc`` to the examples ``CMake`` file.
1191 #. ``suitesparse_test`` is enabled even when ``-DSUITESPARSE=OFF``.
1192 #. ``FixedArray`` internal struct did not respect ``Eigen``
1193 alignment requirements (Koichi Akabe & Stephan Kassemeyer).
1194 #. Fixed ``quadratic.cc`` documentation and code mismatch
1202 #. Fix constant parameter blocks, and other minor fixes (Markus Moll)
1203 #. Fix alignment issues when combining ``Jet`` and
1204 ``FixedArray`` in automatic differeniation.
1205 #. Remove obsolete ``build_defs`` file.
1212 #. Powell's Dogleg solver
1213 #. Documentation now has a brief overview of Trust Region methods and
1214 how the Levenberg-Marquardt and Dogleg methods work.
1218 #. Destructor for ``TrustRegionStrategy`` was not virtual (Markus
1220 #. Invalid ``DCHECK`` in ``suitesparse.cc`` (Markus Moll)
1221 #. Iteration callbacks were not properly invoked (Luis Alberto
1223 #. Logging level changes in ConjugateGradientsSolver
1224 #. VisibilityBasedPreconditioner setup does not account for skipped
1225 camera pairs. This was debugging code.
1226 #. Enable SSE support on MacOS
1227 #. ``system_test`` was taking too long and too much memory (Koichi
1236 #. ``CXSparse`` support.
1237 #. Block oriented fill reducing orderings. This reduces the
1238 factorization time for sparse ``CHOLMOD`` significantly.
1239 #. New Trust region loop with support for multiple trust region step
1240 strategies. Currently only Levenberg-Marquardt is supported, but
1241 this refactoring opens the door for Dog-leg, Stiehaug and others.
1242 #. ``CMake`` file restructuring. Builds in ``Release`` mode by default, and now has platform specific tuning flags.
1243 #. Re-organized documentation. No new content, but better
1250 #. Fixed integer overflow bug in ``block_random_access_sparse_matrix.cc``.
1251 #. Renamed some macros to prevent name conflicts.
1252 #. Fixed incorrent input to ``StateUpdatingCallback``.
1253 #. Fixes to AutoDiff tests.
1254 #. Various internal cleanups.
1262 #. Fix a bug in the handling of constant blocks. (Louis Simard)
1263 #. Add an optional lower bound to the Levenberg-Marquardt regularizer
1264 to prevent oscillating between well and ill posed linear problems.
1265 #. Some internal refactoring and test fixes.
1272 #. New iterative linear solver for general sparse problems - ``CGNR``
1273 and a block Jacobi preconditioner for it.
1274 #. Changed the semantics of how ``SuiteSparse`` dependencies are
1275 checked and used. Now ``SuiteSparse`` is built by default, only if
1276 all of its dependencies are present.
1277 #. Automatic differentiation now supports dynamic number of residuals.
1278 #. Support for writing the linear least squares problems to disk in
1279 text format so that they can loaded into ``MATLAB``.
1280 #. Linear solver results are now checked for nan and infinities.
1281 #. Added ``.gitignore`` file.
1282 #. A better more robust build system.
1287 #. Fixed a strict weak ordering bug in the schur ordering.
1288 #. Grammar and typos in the documents and code comments.
1289 #. Fixed tests which depended on exact equality between floating point
1294 Initial open source release. Nathan Wiegand contributed to the Mac OSX
1301 Ceres Solver grew out of the need for general least squares solving at
1302 Google. In early 2010, Sameer Agarwal and Fredrik Schaffalitzky
1303 started the development of Ceres Solver. Fredrik left Google shortly
1304 thereafter and Keir Mierle stepped in to take his place. After two
1305 years of on-and-off development, Ceres Solver was released as open
1306 source in May of 2012.