Extend nist.cc to test more nonlinear and linear solvers.
(Thanks to Markus Moll for finding the Roszman1 bug)
Change-Id: I92b4bab0771de85f7fe711fb0853f155991f4aaf
1. CostFunction returning false is handled better.
If only the cost is being evaluated, it is possible to
use the false value as an infinite value signal/outside
a region of validity. This allows a weak form of constraint
handling. Useful for example in handling infinities.
2. Changed the way how the slop around zero when model_cost
is larger than the current cost. Relative instead of absolute
tolerances are used. The same logic is propagated how the
corresponding clamping of the model_cost is done.
3. Fixed a minor indexing bug in nist.cc.
4. Some minor logging fixes to nist.cc to make it more
compatible with the rest of ceres.
Together these changes, take the successful solve count from
41/54 to 46/54 and eliminate all NUMERICAL_FAILURE problems.
Change-Id: If94170ea4731af5b243805c0200963dd31aa94a7
Average factorization times for bundle adjustment test problem:
SuiteSparse: 0.2794 s.
CXSparse: 0.4039 s.
CXSparse cached: 0.2399 s.
CXSparse will still be slower, though, because it has to compute
the transpose and J^T * J.
Change-Id: If9cdaa3dd520bee84b56e5fd4953b56a93db6bde
Linking currently fails in Visual Studio due to a missing library
"gomp.lib". This is not needed in Visual Studio. OpenMP works
without it.
Change-Id: I39e204a8dd4f1b7425df7d4b222d86a8bb961432
The overload for pointers in hash tables was applied in normal
usage of schur_ordering.cc. However, the tests did not include the
overload since they only included collections_port.h. As a result,
the routines in schur_ordering.cc were using a different hash
function than that inside the tests.
The fix is to remove the specialization. If this breaks one of the
compiler configurations, we will find a workaround at that time.
Change-Id: Idbf60415d5e2aec0c865b514ad0c577d21b91405
On certain NDK build configurations, one of the innermost
parts of the Schur eliminator would get compiled
incorrectly. The compiler changed a -= to a +=.
The normal Ceres unit tests caught the problem; however,
since it is not possible to build the tests with the NDK
(only with the standalone toolchain) this was difficult to
track down. Finding the issue involved pasting the schur
eliminator unit test inside of solver_impl.cc and other such
hacks.
Change-Id: Ie91bb545d74fe39f0c8cbd1a6eb69ee4d8b25fb2
Eigen3 does not allow column vectors to be stored in row-major
format. NumericDiffCostFunction by default stores its Jacobian
matrices in row-major format. This works fine if the residual
contains more than one variable. But if the residual block
depends on one variable and has more than one residuals, the
resulting Jacobian matrix is a column matrix in row-major format
resulting in a compile time error.
The fix is to check the template parameters and switch to column-major
storage as needed.
Thanks to Lena Gieseke for reporting this.
Change-Id: Icc51c5b38e1f3609e0e1ecb3c4e4a02aecd72c3b
This way, setting the lower and upper bound both to 1.0, one can disable
the automatic trust region scaling.
Change-Id: Ifa317a6911b813a89c1cf7fdfde25af603705319
Headers are installed in ${CMAKE_INSTALL_PREFIX}/include/ceres
Libraries are installed in ${CMAKE_INSTALL_PREFIX}/lib
pdf is installed in ${CMAKE_INSTALL_PREFIX}/share/ceres/docs
Change-Id: Ic175f2c2f5fa86820a1e8c64c2ed171f4a302a68
If the source directory is a clone, at CMake time the commit-msg hook gets
downloaded and installed in the right location.
Change-Id: I5fee17d050ca22d8b92a49fdcc2a1cd6659f209b
Currently the examples are always built. For external projects, it is useful
not to compile the examples.
Change-Id: I41d3bde19c7e742818e60f78222d39c43992ca8b
For problems with a small number of variables, but a large
number of residuals, it is sometimes beneficial to use the
Cholesky factorization on the normal equations, instead of
the dense QR factorization of the Jacobian, even though it
is numerically the better thing to do.
Change-Id: I3506b006195754018deec964e6e190b7e8c9ac8f
Gather
* all executables in ${CMAKE_BINARY_DIR}/bin
* all libraries (static and dynamic) in ${CMAKE_BINARY_DIR}/lib
Change-Id: Ibc2fa1adfb6f0aea65d66d570259b79546bf3b07
SuiteSparse 4 requires linking to libsuitesparseconfig.a.
Both SuiteSparse 3 and SuiteSparse 4 require an additional header
(either UFconfig.h or SuiteSparse_config.h) that is not found if it is
in a separate path. Therefore, add explicit checks.
Change-Id: I699902b5db4f1b7f17134b5a54f9aa681445e294
In the TrustRegionMinimizer, the step is currently implicitly negated.
This is done so that the linearized residual is |r - J*step|^2, which
corresponds to J*step = r, so neither J nor r have to be modified.
However, it leads to the rather unintuitive situation that the strategy
returns a step in positive gradient direction, which you would expect to
increase the function value. One way is to rename the "step" parameter in
the strategy to "negative_step" and document it.
This patch instead moves the negation inside the strategy, just around
the linear solver call, so that it is done in a local context and easier
to document.
Change-Id: Idb258149a01f61c64e22128ea221c5a30cd89c89
Sameer Agarwal