Previously, even when an iteration was successful, the LineSearchMinimizer
would only add the iteration summary to the Summary object if none
of the convergence tests were passed. This could cause iterations with
significant progress in the last iteration to be mis-reported.
The solution would be correct, but the actual cost would be misreported.
This change changes the order of these operations and ensures that
the iteration summary is added whenever the iteration itself is successful.
Thanks to Daniel Weindl for reporting this.
Change-Id: Iff10eccb49d50ad28127f44e149c17fa466db4ae
- When compiling on Clang on OS X with MINIGLOG enabled, name lookup
finds two ambiguous definitions of WARNING, one in the global
namespace and one in the google namespace, both defined in the
miniglog version of logging.h.
Change-Id: I6f1ad7d2750e1ed20ec1ba4574aab599086431df
1. Add schur_templates.cc to Android.mk
2. When detecting the Schur structure of the Jacobian,
the check for whether linear solver being used is indeed
of Schur type or not, should use the LinearSolver::Options
struct created and populated by the preprocessor rather than
depending on the value in the input Solver::Options. The reason
is that the preprocessor may change the linear solver type
depending on the lack of a Schur structure in the problem.
Change-Id: I6f018f6817c05d704409181c7b1e25155528ab84
1. Refactor the python code that generates the template specializations
to remove code duplication.
2. Improved the logic for template specialization selection where
Eigen::Dynamic now serves as a wildcard.
3. Added schur_templates.h/cc which allows querying the set of available
template specializations without instantiating a linear solver.
4. Added Solver::Summary::schur_structre_given and
Solver::Summary::schur_structure_used and expose them in
Solver::Summary::FullReport for better performance debugging.
5. Updated the templates with newer dates and some minor comments cleanup
which lead to the the template specializations to be re-generated.
Change-Id: Iaf3c6f714353597899916c300465da01f151c3de
- ScalarBinaryOpTraits was introduced in Eigen 3.3, and thus breaks
compilation of Ceres for Eigen versions < 3.3.
- Reported as issue #252 by AustinDeric:
https://github.com/ceres-solver/ceres-solver/issues/252
Change-Id: Ie9b7dace17709c348b18b297f2d5306b054ce4b0
Changes in Eigen's implementation for FullPivLU significantly
degraded the performance of the line search as reported by Weiguang.
We do not completely understand what is going on, as Eigen's changes
seem sane. So for now, this change explicitly works around the
changes made by Eigen to restore the performance of the line search.
Figuring out the underlying problem and fixing it remains an open
issue.
https://github.com/ceres-solver/ceres-solver/issues/248
Change-Id: I9993d73a09dc990ab567ce6bc447f16eac74abec
This commit extends the use of Jets with Eigen matrices and arrays
by enabling the use of binary operators (such as scalar addition,
multiplication, as well as matrix multiplication) when one Eigen
matrix/array is of type Jet and the other is a scalar type. This
should increase performance since Jet types have optimized
scalar-to-jet binary operations.
Change-Id: Ia756064845d845cefcf0abb16d366331d2824b52
Binary operations between Jets and doubles are well defined
and should not require an explicit conversion to Jets to work.
This was an oversight earlier and lead to overzealous conversions
all over our in our example code.
Change-Id: I1799770818e136edfc0a5802d86037ce9aec4923
ResidualBlock evaluation has logic to ensure that CostFunction
should always fill out the residual and jacobian arrays completely
by using a special value to pre-populate these arrays.
This works for CostFunctions with analytical Jacobians but not for
AutoDiffCostFunction and NumericDiffCostFunction Jacobians.
There is no way to fix this for NumericDiffCostFunctions without
introducing significant performance penalties but the residual
evaluation fails, which should be enough to catch such errors.
For AutoDiffCostFunction the way the Jets are default initialized
was sidestepping this check. So now, the Jet that is used to
capture the output residuals is now initialized with
kImpossibleValue, which will ensure that if the user forgets
to fill all output fields, it triggers an evaluation error.
This change required that ceres::internal::kImpossibleValue be moved
out of array_utils.h/cc to types.h.
Change-Id: I35bb0946cf0785a5d43c7b5459a2272848fb2a9b
MSVC requires using google::GLOG_WARNING, but this breaks the build
inside google. So add a MSVC specific ifdef to guard against it.
Change-Id: I3509bb3fd2c01722a1c76545b57d2bf0224f6d6e
- Previously we were not listing gflags as a public dependency of Ceres
if it and glog were found (and MINIGLOG was not being used). This
does not reflect that if glog was compiled with gflags then it will
#include gflags/gflags.h in glog/logging.h, thus making gflags a
public dependency of anything linking against glog.
- On *nix OSs if glog/gflags are shared libraries this did not result
in a link error when compiling Ceres as the gflags symbols were
indirectly resolved. However, on MSVC this is not the case, and this
could result in unresolved gflags symbol link errors when compiling
Ceres.
- Now we add gflags to the list of public Ceres dependencies if both
glog and gflags are found (and MINIGLOG is not enabled).
Change-Id: I5ce6038fa816781cc81b378522068dc563d29c51
- GLOG_NO_ABBREVIATED_SEVERITIES is the default on Windows, in which
case google::WARNING is not defined.
- Remove Ceres-specific redefinition of WARNING in mock-log.h fork
in place of using non-abbreviated severity in the one place in
levenberg_marquardt_strategy_test where ScopedMockLog is actually
used.
- Remove unnecessary dependency of gradient_checking_cost_function_test
on ScopedMockLog.
Change-Id: I9fb540f638037b6015fd264cfc618c9d60f5686c
Support for parameter tolerance was added to the line search
minimizer was added a while ago, and calling Solve on a
non-linear least squares problem supported it but for some reason
the GradientProblemSolver::Options struct was missing this
option even though the documentation suggested that it was present!
Thanks to Noah Snavely for reporting this bug.
Change-Id: I57cf4ab396bc822c19fa298529e113b89664a349
This commit relaxes the tolerance value for comparing between the actual
local matrix and the expected local matrix. Without this fix,
EigenQuaternionParameterization.ZeroTest could fail as the difference
exactly matches the value of std::numeric_limits<double>::epsilon().
Change-Id: Ic4d3f26c0acdf5f16fead80dfdc53df9e7dabbf9
Previously, the test for the projective camera model would fail as no
tolerance is set in line 144. To resolve this, this commit changes
assert_equal to assert_near.
Change-Id: I6cd3379083b1a10c7cd0a9cc83fd6962bb993cc9
Any result of an arithmetic operation on floating-point matrices
should never be checked for strict equality with some expected
value, due to limited floating point precision on different machines.
This fixes some occurences of exact checks in the gradient checker
unit test that were causing problems on some platforms.
Change-Id: I48e804c9c705dc485ce74ddfe51037d4957c8fcb
Intel C compiler strictly asks for parallel loops with collapse to be
perfectly nested. Otherwise, compiling Ceres with ICC will throw an
error at line 348 of covariance_impl.cc.
Change-Id: I1ecb68e89b7faf79e4153dfe6675c390d1780db4
1. SubsetParameterization can now be constructed such that all
parameters are constant. This is required for it be used as part
of a ProductParameterization to hold a part of parameter block
constant. For example, a parameter block consisting of a rotation
as a quaternion and a translation vector can now have a local
parameterization where the translation part is constant and the
quaternion part has a QuaternionParameterization associated with it.
2. The check for the tangent space of a parameterization being
positive dimensional. We were not doing this check up till now
and the user could accidentally create parameterizations like this
and create a problem for themselves. This will ensure that even
though one can construct a SubsetParameterization where all
parameters are constant, you cannot actually use it as a local
parameterization for an entire parameter block. Which is how
it was before, but the check was inside the SubsetParameterization
constructor.
3. Added more tests and refactored existing tests to be more
granular.
Change-Id: Ic0184a1f30e3bd8a416b02341781a9d98e855ff7
The documentation for ExpectArraysClose and its implementation
did not match.
This change makes the polynomial_test not fail on 64bit AMD builds.
Thanks to Phillip Huebner for reporting this.
Change-Id: I503f2d3317a28d5885a34f8bdbccd49d20ae9ba2
Returning by reference leads to lifetime issues with the default
value which may go out of scope by the time it is used.
Thanks to @Ardavel for reporting this, as this causes graph_test
to fail on VS2015x64.
https://github.com/ceres-solver/ceres-solver/issues/216
Change-Id: I596481219cfbf7622d49a6511ea29193b82c8ba3
This is a follow-up on c/7470. GradientCheckingCostFunction calls
callback_->SetGradientErrorDetected() in its Evaluate method,
which will run in multiple threads simultaneously when enabling
this option in the solver. Thus, the string append operation
inside that method has to be protected by a mutex.
Change-Id: I314ef1df2be52595370d9af05851bf6da39bb45e
Solver::Options::numeric_derivative_relative_step_size to
Solver::Options::gradient_check_numeric_derivative_relative_step_size
Change-Id: Ib89ae3f87e588d4aba2a75361770d2cec26f07aa
- Previously we disabled OpenMP if Clang was detected, as it did not
support it. However as of Clang 3.8 (and potentially Xcode 8) OpenMP
is supported.
Change-Id: Ia39dac9fe746f1fc6310e08553f85f3c37349707
Change the Ceres gradient checking API to make is useful for
unit testing, clean up code duplication and fix interaction between
gradient checking and local parameterizations.
There were two gradient checking implementations, one being used
when using the check_gradients flag in the Solver, the other
being a standalone class. The standalone version was restricted
to cost functions with fixed parameter sizes at compile time, which
is being lifted here. This enables it to be used inside the
GradientCheckingCostFunction as well.
In addition, this installs new hooks in the Solver to ensure
that Solve will fail if any incorrect gradients are detected. This
way, you can set the check_gradient flags to true and detect
errors in an automated way, instead of just printing error information
to the log. The error log is now also returned in the Solver summary
instead of being printed directly. The user can then decide what to
do with it. The existing hooks for user callbacks are used for
this purpose to keep the internal API changes minimal and non-invasive.
The last and biggest change is the way the the interaction between
local parameterizations and the gradient checker works. Before,
local parameterizations would be ignored by the checker. However,
if a cost function does not compute its Jacobian along the null
space of the local parameterization, this wil not have any effect
on the solver, but would result in a gradient checker error.
With this change, the Jacobians are multiplied by the Jacobians
of the respective local parameterization and thus being compared
in the tangent space only.
The typical use case for this are quaternion parameters, where
a cost function will typically assume that the quaternion is
always normalized, skipping the correct computation of the Jacobian
along the normal to save computation cost.
Change-Id: I5e1bb97b8a899436cea25101efe5011b0bb13282
TrustRegionMinimizer evaluates the size of the step
taken in the ambient space, where as the LineSearchMinimizer
was using the norm in the tangent space. This change fixes
this discrepancy.
Change-Id: I9fef64cbb5622c9769c0413003cfb1dc6e89cfa3
- Previously we were auto-detecting a "64" suffix for the install path
for the Ceres library on non-Debian/Arch Linux distributions, but
we were installing CeresConfig.cmake to an architecture independent
location.
- We now install CeresConfig.cmake to lib${LIB_SUFFIX}/cmake/Ceres.
- Also make LIB_SUFFIX visible to the user in the CMake GUI s/t they can
easily override the auto-detected value if desired.
- Reported by jpgr87@gmail.com as Issue #194.
Change-Id: If126260d7af685779487c01220ae178ac31f7aea
The norm comparison in QuaternionParameterizationHelper is exact
but that leads to numerical precision problems and test failures
as reported by Nicu Stiurca.
https://github.com/ceres-solver/ceres-solver/issues/198
Change-Id: I45f42f0373770408ddeee0b4c9d162e619f8d445
on a ceres built without SuiteSparse support.
This is only a minor issue, up until now this case would have been handled by
the "default" case of the switch
Change-Id: Ib4568eb827dbf0e8e8d618a5bc8cd379a3c42c42
This variable should be initialized before the decision to
perform inner iterations are performed.
Change-Id: Ic80e2d5284ecc67eaed1859be009b1a4d18bf76b
1. Break up the monolithic loop in TrustRegionMinimizer::Minimize
into a number of more easily described and analyzed subfunctions.
2. Break out the logic for evaluating the quality of a Trust Region
step into its own object - TrustRegionStepEvaluator.
Change-Id: I08580ecac074cfd74c096cb8e4880cbda3d48296
Sameer Agarwal