This adds a callback mechanism to for users to get notified just
before jacobian and residual evaluations. This will enable
aggressive caching and sharing of compute between cost functions.
Change-Id: I67993726920218edf71ab9ae70c34c204756c71a
1. Solver::Options::num_threads now controls parallelism in Ceres
Solver. The user specified value of
Solver::Options::num_linear_solver_threads is ignored.
2. If the user specifies Solver::Options::num_linear_solver_threads
and it is different from Solver::Options::num_threads,
a warning is printed.
3. Solver::Summary:num_linear_solver_threads_given and
Solver::Summary::num_linear_solver_threads_used are also
deprecated and are always set to Solver::Summary::num_threads_given
and Solver::Summary::num_threads_used.
Change-Id: I20b9336d9336e400e6f0a15b63857c0c43eb271c
Fixes a bug introduced in 12280 that incorrectly used xor for multiple
values.
Tested it manually by verifying it would fail for 2, 3, or 4 of them
defined.
Change-Id: I90c2662f4744aceaf0119dbbe072beced3f67c9a
Implements ParallelFor using the C++11 based ThreadPool. The C++11
parallel for is 50-70% faster than single threaded, and 20-30% slower
than TBB.
Tested by compiling with OpenMP, TBB, and C++11 Threading support and
ran the unit tests. Ran bazel as well.
Change-Id: I7fd6c9037ff9f200ce6999b5f39918995bb6b8ea
A Ceres Context holds common global state that can be re-used within
Ceres. The Context current contains a thread pool if compiling with
C++11 threading support. Threads are expensive to create and destroy so
it is good to maintain across multiple Ceres solves.
Tested by compiling with and without TBB support and ran unit tests. Ran
bazel as well.
Change-Id: I82f598dfae642aa0e81a6039dc174608a5e8dbfb
1. Replace two maps by 1.
2. Update number of calls and the time for the call at the cost
of a single map lookup.
3. Add Solver::Summary::num_linear_solves.
Fixes https://github.com/ceres-solver/ceres-solver/issues/340
Change-Id: I71eb9be7fb363a8cb066591c4c1761f256c81677
1. Fix a typo in auto_diff_cost_function.h
2. Fix and update Solver::Summary::FullReport() text labels.
3. Add logging of the number of residual and jacobian evaluations
to the full report. The GradientProblemSolver already does this.
Change-Id: I41059af5f0ebe0417accbbc30b0808a4b04b9edb
Most matrices in Ceres are row-major. Eigen does now
allow statically sized row-major matrices with a single
column.
This CL changes the EigenTypes struct to change the definition
of typedefs to change to ColMajor if this is the case.
This will allow Schur specializations (2,1,6) for example.
Change-Id: I24e7b62d4e1cb5660541062336225bbc9c41c6bf
The non-linear least squares solver had the ability to update the
user's parameters every iteration. Now GradientProblemSolver can
do the same.
Also a few minor Sphinx markup related cleanup to the documentation
which were found in the process of updating Sphinx on my machine
and adding the docs for this feature.
This fixes https://github.com/ceres-solver/ceres-solver/issues/246
Change-Id: Ib6b90ac22be8bfb60b14f25ad52082ba371af164
Would result in compilation error
include/ceres/jet.h:296:59: error: requested alignment is not a positive power of 2
alignas(kAlignment) Eigen::Matrix<T, N, 1, kAlignOrNot> v;
because CERES_JET_ALIGN_BYTES = EIGEN_MAX_ALIGN_BYTES = 0 in that case.
Change-Id: I247653d3b7555e80891d3fb78676acfa302a9e1d
- When compiling Ceres without C++11 but with glog, then the unnamed
enum results in a compiler warning when
google::GetReferenceableValue(.) is invoked with the unnamed enum as a
template parameter (unnamed-type-template-args) which was not legal
prior to C++11.
Change-Id: Iddbed05e2199865ab915f2c5817ce917bb0609b5
1. Default linear solver is Eigen::LDLT
2. Options::max_iterations -> Options::max_num_iterations
3. Options::error_threshold -> Options::cost_threshold
4. Options::relative_step_threshold -> Options::parameter_threshold
5. Options::initial_scale_factor -> Options::initial_trust_region_radius
6. The default values of the above parameters have been changed
to match those in ceres::Solver::Options
7. Status::RUNNING has been removed
8. Update now returns a bool instead of a Status enum and
the status handling has been included in the main loop.
9. Summary::gradient_norm has been changed to Summary::gradient_max_norm
to match the convergence test
10. A member variable cost_ has been added which is computed by Update
11. The test for parameter_tolerance based convergence is made
more robust near zero.
12. Use of double has been replaced by Scalar.
13. Minor clang-formatting
Change-Id: I3cb0e2fd0a0204476bb8718761dc740cdf5e42ce
The solver code must rely on the vectors for
sizing, since not all cost functions will have
NumParameters() or NumResiduals().
Change-Id: Id254ce37507443910edb0064de7907d64558851e
1. Default constructor and initialization for Summary.
2. Add Jacobi scaling.
3. Add bounds on the lm diagonal
4. Use the diagonal of J'J as the regularizer instead of identity.
5. Update the computation of rho to match the change in regularization.
As a result of these changes, the performance of TinySolver is
now the same as ceres::Solver, solving 53 out of 54 problems.
Change-Id: Ie08c3389ac2e3964ffa04411734c06b65835358a
1. Instead of Core/LU just include Eigen/Dense
2. Rename SolverParameters to Options and params to options.
3. Rename Results to Summary.
4. Summary::error_magnitude -> Summary::final_cost.
5. Add Summary::initial_cost.
6. Change definitions of Summary::initial_cost and Summary::final_cost
to match those used by Ceres::Solver.
Change-Id: Id64b78398f47810ca25938a15423c514fc8c164d
1. Change the ordering from NUM_PARAMETERS, NUM_RESIDUALS to
NUM_RESIDUALS, NUM_PARAMETERS in docs and in code.
2. TinySolver::solve -> TinySolver::Solve
Change-Id: I4dca87b971fd9168f1200b53c362669cffc82c1b
Tiny solver is targeted towards small dense least square
solves, where the overhead of calling normal Ceres is too
high. For example, when solving for inverse camera
distortion for every pixel location in a many-megapixel
image. Anecdotally, at one point in the past, tiny solver
was ~20x faster than Ceres for the problems it's intended
for. This is due to two key aspects:
1. Memory is allocated up front: repeated solves incur no
allocation overhead beyond a few scalars on the stack.
2. The cost function is fully inlined into the solver
loop, removing even the cost function call overhead.
Tiny solver originated many years ago as part of
libmv/Blender, where it is still used for distortion solving
today, but the time has come for it to migrate into Ceres.
This commit is just the initial import into Ceres. Follow
up patches will add further cleanups, and add CostFunction
and Jet adapters to make it easier to call tiny solver
(though by using adapters, some performance advantages will
be lost).
Change-Id: I8079535cd41382b1e0ac0ca2fca141711c72b7f8
1. Fix a bug which was causing the cost and gradient evaluation
time to not be reported.
2. Add the number of times cost and gradients are evaluated to
the Summary object and to the output of FullReport.
Change-Id: Id0703cd2dafbf437f3e537fbdc30ae81d5f4f540
THIS IS AN API BREAKING CHANGE.
Decouple the algorithm from the sparse linear algebra
library being used to perform the computation.
Before this change
Covariance::AlgorithmType had values
DENSE_SVD
EIGEN_SPARSE_QR
SUITE_SPARSE_QR
This has been replaced by two enums now.
Covariance::Options::sparse_linear_algebra_library_type
which can take values EIGEN_SPARSE, SUITE_SPARSE or CX_SPARSE.
The last one is currently not supported.
And Covariance::Options::algorithm_type takes values
DENSE_SVD
SPARSE_QR
This sets the stage for future extensions of the covariance
computation algorithm.
Also as part of this change, the covariance computation chapter
has been made a top level chapter on its own instead of being
buried deep inside the Solving Non-linear Least Squares problem.
Change-Id: Ibfbf60902d8d17694d9ff585047a5a57d329ab22
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
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
Below fixes are required to compile Ceres as a dynamic library using
MSVC.
1. Added CERES_EXPORT to EigenQuaternionParameterization.
2. Removed CERES_EXPORT from CubicInterpolator and BicubicInterpolator.
EigenQuaternionParameterization has its source embedded in ceres.dll
whereas Cubic+BicubicInterpolators are defined in the header file and
therefore do not require __declspec(dllimport).
Change-Id: Idd1f387890c36aab51a1a4177ea7c06933b96f39
- MSVC deprecated the standard POSIX names for Bessel functions in
favour of underscore prefixed versions.
- Previously we were checking for the presence of the newer underscore
functions via a macro, which fails as the functions are not defined
as macros.
- Now we check via a check_cxx_source_compiles() if the newer underscore
prefixed versions exist when compiling on MSVC and use a new Ceres
configuration #define to switch on their presence in jet.h
Change-Id: I430880bde2981d12f4d03dbc94d903b9842e887e
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
- Eigen versions < 3.3 only supported SIMD instructions that required
16-byte alignment (SSE), whereas Eigen >= 3.3 also supports AVX
instructions which require 32+-byte alignment.
- Previously we only ever requested 16-byte alignment for Jets (if >=
C++11 was enabled) which resulted in Eigen assertions being triggered
on some compilers as reported as Issue #251:
https://github.com/ceres-solver/ceres-solver/issues/251.
- Now we use Eigen’s EIGEN_MAX_ALIGN_BYTES macro, defined in Eigen >=
3.3 to specify the byte alignment for Jets when C++11 is enabled for
Eigen >= 3.3. For Eigen versions < 3.3, we maintain the previous
behaviour of 16.
Change-Id: I749af7a70ae794e0c2a59301128db781e338422c
- alignas(0) should be ignored, however it results in a build error on
GCC, so instead default to the alignment of double in Jets if
we cannot align to 16-byte boundaries on the platform, but are
compiling with C++11.
Change-Id: I2e54c69516ea2e1447a8bdc138b2dd70050c6dad
- As per Andrew Hunter’s comments in the commit which added Jet
alignment when using C++11 here:
https://ceres-solver-review.googlesource.com/#/c/7100, there is wide
lattitude in the standard about what the maximum supported alignment
can be.
- Previously, we were forcing the alignment to 1, if the value of
alignof(std::max_align_t), which we use as a proxy for the maximum
supported alignment on the platform, was < 16.
- An alignment of 1 is not valid for Jets, as it would weaken the
natural alignment of the types within a Jet, which would typically be
4 (32-bit systems) or 8 (64-bit systems), thus resulting in a compiler
error.
- This was reported as issue 235 for Clang 3.8 on i386:
https://github.com/ceres-solver/ceres-solver/issues/235.
Change-Id: Ie39e5499c64f9231f29ebf4392992b5c9ce2e385
In the process also move some documentation from the file header to
just above the class declaration.
This change is in response to https://github.com/ceres-solver/ceres-solver/issues/233
Change-Id: I08cff1a94c57b67dd2bd8be4dba8c3fce46b68ab
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
They were present as debugging checks but were causing problems
with the build on 32bit i386 due to numerical cancellation issues,
where x ~ -epsilon.
Removing these checks only changes the behaviour in Debug mode.
We are already handling such small negative numbers in production
if they occur. All that this change does is to remove the crash.
https://github.com/ceres-solver/ceres-solver/issues/212
Thanks to @NeroBurner and @debalance for reporting this.
Change-Id: I66480e86d4fa0a4b621204f2ff44cc3ff8d01c04
Keir Mierle