Installation¶

Getting the source code¶

git clone https://ceres-solver.googlesource.com/ceres-solver


Dependencies¶

Note

Starting with v2.0 Ceres requires a fully C++14-compliant compiler. In versions <= 1.14, C++11 was an optional requirement.

Ceres relies on a number of open source libraries, some of which are optional. For details on customizing the build process, see Customizing the build .

• Eigen 3.3 or later required.

Note

Ceres can also use Eigen as a sparse linear algebra library. Please see the documentation for EIGENSPARSE for more details.

• CMake 3.5 or later required.

• glog 0.3.1 or later. Recommended

glog is used extensively throughout Ceres for logging detailed information about memory allocations and time consumed in various parts of the solve, internal error conditions etc. The Ceres developers use it extensively to observe and analyze Ceres’s performance. glog allows you to control its behaviour from the command line. Starting with -logtostderr you can add -v=N for increasing values of N to get more and more verbose and detailed information about Ceres internals.

Ceres also ships with a minimal replacement of glog called miniglog that can be enabled with the MINIGLOG build option. miniglog is supplied for platforms which do not support the full version of glog.

In an attempt to reduce dependencies, it may be tempting to use miniglog on platforms which already support glog. While there is nothing preventing the user from doing so, we strongly recommend against it. miniglog has worse performance than glog and is much harder to control and use.

Note

If you are compiling glog from source, please note that currently, the unit tests for glog (which are enabled by default) do not compile against a default build of gflags 2.1 as the gflags namespace changed from google:: to gflags::. A patch to fix this is available from here.

• gflags. Needed to build examples and tests and usually a dependency for glog.

• SuiteSparse. Needed for solving large sparse linear systems. Optional; strongly recomended for large scale bundle adjustment

Note

If SuiteSparseQR is found, Ceres attempts to find the Intel Thread Building Blocks (TBB) library. If found, Ceres assumes SuiteSparseQR was compiled with TBB support and will link to the found TBB version. You can customize the searched TBB location with the TBB_ROOT variable.

• CXSparse. Similar to SuiteSparse but simpler and slower. CXSparse has no dependencies on LAPACK and BLAS. This makes for a simpler build process and a smaller binary. Optional

• Apple’s Accelerate sparse solvers. As of Xcode 9.0, Apple’s Accelerate framework includes support for solving sparse linear systems across macOS, iOS et al. Optional

• BLAS and LAPACK routines are needed by SuiteSparse, and optionally used by Ceres directly for some operations.

On UNIX OSes other than macOS we recommend ATLAS, which includes BLAS and LAPACK routines. It is also possible to use OpenBLAS . However, one needs to be careful to turn off the threading inside OpenBLAS as it conflicts with use of threads in Ceres.

MacOS ships with an optimized LAPACK and BLAS implementation as part of the Accelerate framework. The Ceres build system will automatically detect and use it.

For Windows things are much more complicated. LAPACK For Windows has detailed instructions..

Optional but required for SuiteSparse.

Linux¶

We will use Ubuntu as our example linux distribution.

Note

These instructions are for Ubuntu 18.04 and newer. On Ubuntu 16.04 you need to manually get a more recent version of Eigen, such as 3.3.7.

Start by installing all the dependencies.

# CMake
sudo apt-get install cmake
# BLAS & LAPACK
sudo apt-get install libatlas-base-dev
# Eigen3
sudo apt-get install libeigen3-dev
# SuiteSparse and CXSparse (optional)
sudo apt-get install libsuitesparse-dev


We are now ready to build, test, and install Ceres.

tar zxf ceres-solver-2.0.0.tar.gz
mkdir ceres-bin
cd ceres-bin
cmake ../ceres-solver-2.0.0
make -j3
make test
# Optionally install Ceres, it can also be exported using CMake which
# allows Ceres to be used without requiring installation, see the documentation
make install


You can also try running the command line bundling application with one of the included problems, which comes from the University of Washington’s BAL dataset [Agarwal].

bin/simple_bundle_adjuster ../ceres-solver-2.0.0/data/problem-16-22106-pre.txt


This runs Ceres for a maximum of 10 iterations using the DENSE_SCHUR linear solver. The output should look something like this.

iter      cost      cost_change  |gradient|   |step|    tr_ratio  tr_radius  ls_iter  iter_time  total_time
0  4.185660e+06    0.00e+00    1.09e+08   0.00e+00   0.00e+00  1.00e+04       0    7.59e-02    3.37e-01
1  1.062590e+05    4.08e+06    8.99e+06   5.36e+02   9.82e-01  3.00e+04       1    1.65e-01    5.03e-01
2  4.992817e+04    5.63e+04    8.32e+06   3.19e+02   6.52e-01  3.09e+04       1    1.45e-01    6.48e-01
3  1.899774e+04    3.09e+04    1.60e+06   1.24e+02   9.77e-01  9.26e+04       1    1.43e-01    7.92e-01
4  1.808729e+04    9.10e+02    3.97e+05   6.39e+01   9.51e-01  2.78e+05       1    1.45e-01    9.36e-01
5  1.803399e+04    5.33e+01    1.48e+04   1.23e+01   9.99e-01  8.33e+05       1    1.45e-01    1.08e+00
6  1.803390e+04    9.02e-02    6.35e+01   8.00e-01   1.00e+00  2.50e+06       1    1.50e-01    1.23e+00

Ceres Solver v2.0.0 Solve Report
----------------------------------
Original                  Reduced
Parameter blocks                        22122                    22122
Parameters                              66462                    66462
Residual blocks                         83718                    83718
Residual                               167436                   167436

Minimizer                        TRUST_REGION

Dense linear algebra library            EIGEN
Trust region strategy     LEVENBERG_MARQUARDT

Given                     Used
Linear solver                     DENSE_SCHUR              DENSE_SCHUR
Linear solver ordering              AUTOMATIC                22106, 16

Cost:
Initial                          4.185660e+06
Final                            1.803390e+04
Change                           4.167626e+06

Minimizer iterations                        6
Successful steps                            6
Unsuccessful steps                          0

Time (in seconds):
Preprocessor                            0.261

Residual evaluation                   0.082
Jacobian evaluation                   0.412
Linear solver                         0.442
Minimizer                               1.051

Postprocessor                           0.002
Total                                   1.357

Termination:                      CONVERGENCE (Function tolerance reached. |cost_change|/cost: 1.769766e-09 <= 1.000000e-06)


macOS¶

On macOS, you can either use Homebrew (recommended) or MacPorts to install Ceres Solver.

If using Homebrew, then

brew install ceres-solver


will install the latest stable version along with all the required dependencies and

brew install ceres-solver --HEAD


If using MacPorts, then

sudo port install ceres-solver


You can also install each of the dependencies by hand using Homebrew. There is no need to install BLAS or LAPACK separately as macOS ships with optimized BLAS and LAPACK routines as part of the vecLib framework.

# CMake
brew install cmake
brew install glog gflags
# Eigen3
brew install eigen
# SuiteSparse and CXSparse
brew install suite-sparse


We are now ready to build, test, and install Ceres.

tar zxf ceres-solver-2.0.0.tar.gz
mkdir ceres-bin
cd ceres-bin
cmake ../ceres-solver-2.0.0
make -j3
make test
# Optionally install Ceres, it can also be exported using CMake which
# allows Ceres to be used without requiring installation, see the
make install


Building with OpenMP on macOS¶

Up to at least Xcode 12, OpenMP support was disabled in Apple’s version of Clang. However, you can install the latest version of the LLVM toolchain from Homebrew which does support OpenMP, and thus build Ceres with OpenMP support on macOS. To do this, you must install llvm via Homebrew:

# Install latest version of LLVM toolchain.
brew install llvm


As the LLVM formula in Homebrew is keg-only, it will not be installed to /usr/local to avoid conflicts with the standard Apple LLVM toolchain. To build Ceres with the Homebrew LLVM toolchain you should do the following:

tar zxf ceres-solver-2.0.0.tar.gz
mkdir ceres-bin
cd ceres-bin
# Configure the local shell only (not persistent) to use the Homebrew LLVM
# toolchain in favour of the default Apple version.  This is taken
# verbatim from the instructions output by Homebrew when installing the
# llvm formula.
export LDFLAGS="-L/usr/local/opt/llvm/lib -Wl,-rpath,/usr/local/opt/llvm/lib"
export CPPFLAGS="-I/usr/local/opt/llvm/include"
export PATH="/usr/local/opt/llvm/bin:$PATH" # Force CMake to use the Homebrew version of Clang and enable OpenMP. cmake -DCMAKE_C_COMPILER=/usr/local/opt/llvm/bin/clang -DCMAKE_CXX_COMPILER=/usr/local/opt/llvm/bin/clang++ -DCERES_THREADING_MODEL=OPENMP ../ceres-solver-2.0.0 make -j3 make test # Optionally install Ceres. It can also be exported using CMake which # allows Ceres to be used without requiring installation. See the # documentation for the EXPORT_BUILD_DIR option for more information. make install  Like the Linux build, you should now be able to run bin/simple_bundle_adjuster. Windows¶ Note If you find the following CMake difficult to set up, then you may be interested in a Microsoft Visual Studio wrapper for Ceres Solver by Tal Ben-Nun. On Windows, we support building with Visual Studio 2015.2 of newer. Note that the Windows port is less featureful and less tested than the Linux or macOS versions due to the lack of an officially supported way of building SuiteSparse and CXSparse. There are however a number of unofficial ways of building these libraries. Building on Windows also a bit more involved since there is no automated way to install dependencies. Note Using google-glog & miniglog with windows.h. The windows.h header if used with GDI (Graphics Device Interface) defines ERROR, which conflicts with the definition of ERROR as a LogSeverity level in google-glog and miniglog. There are at least two possible fixes to this problem: 1. Use google-glog and define GLOG_NO_ABBREVIATED_SEVERITIES when building Ceres and your own project, as documented here. Note that this fix will not work for miniglog, but use of miniglog is strongly discouraged on any platform for which google-glog is available (which includes Windows). 2. If you do not require GDI, then define NOGDI before including windows.h. This solution should work for both google-glog and miniglog and is documented for google-glog here. 1. Make a toplevel directory for deps & build & src somewhere: ceres/ 2. Get dependencies; unpack them as subdirectories in ceres/ (ceres/eigen, ceres/glog, etc) 1. Eigen 3.3 . Configure and optionally install Eigen. It should be exported into the CMake package registry by default as part of the configure stage so installation should not be necessary. 2. google-glog Open up the Visual Studio solution and build it. 3. gflags Open up the Visual Studio solution and build it. 4. (Experimental) SuiteSparse Previously SuiteSparse was not available on Windows, recently it has become possible to build it on Windows using the suitesparse-metis-for-windows project. If you wish to use SuiteSparse, follow their instructions for obtaining and building it. 5. (Experimental) CXSparse Previously CXSparse was not available on Windows, there are now several ports that enable it to be, including: [1] and [2]. If you wish to use CXSparse, follow their instructions for obtaining and building it. 3. Unpack the Ceres tarball into ceres. For the tarball, you should get a directory inside ceres similar to ceres-solver-2.0.0. Alternately, checkout Ceres via git to get ceres-solver.git inside ceres. 4. Install CMake, 5. Make a dir ceres/ceres-bin (for an out-of-tree build) 6. Run CMake; select the ceres-solver-X.Y.Z or ceres-solver.git directory for the CMake file. Then select the ceres-bin for the build dir. 7. Try running Configure. It won’t work. It’ll show a bunch of options. You’ll need to set: 1. Eigen3_DIR (Set to directory containing Eigen3Config.cmake) 2. GLOG_INCLUDE_DIR_HINTS 3. GLOG_LIBRARY_DIR_HINTS 4. (Optional) gflags_DIR (Set to directory containing gflags-config.cmake) 5. (Optional) SUITESPARSE_INCLUDE_DIR_HINTS 6. (Optional) SUITESPARSE_LIBRARY_DIR_HINTS 7. (Optional) CXSPARSE_INCLUDE_DIR_HINTS 8. (Optional) CXSPARSE_LIBRARY_DIR_HINTS to the appropriate directories where you unpacked/built them. If any of the variables are not visible in the CMake GUI, create a new entry for them. We recommend using the <NAME>_(INCLUDE/LIBRARY)_DIR_HINTS variables rather than setting the <NAME>_INCLUDE_DIR & <NAME>_LIBRARY variables directly to keep all of the validity checking, and to avoid having to specify the library files manually. 8. You may have to tweak some more settings to generate a MSVC project. After each adjustment, try pressing Configure & Generate until it generates successfully. 9. Open the solution and build it in MSVC To run the tests, select the RUN_TESTS target and hit Build RUN_TESTS from the build menu. Like the Linux build, you should now be able to run bin/simple_bundle_adjuster. Notes: 1. The default build is Debug; consider switching it to release mode. 2. Currently system_test is not working properly. 3. CMake puts the resulting test binaries in ceres-bin/examples/Debug by default. 4. The solvers supported on Windows are DENSE_QR, DENSE_SCHUR, CGNR, and ITERATIVE_SCHUR. 5. We’re looking for someone to work with upstream SuiteSparse to port their build system to something sane like CMake, and get a fully supported Windows port. Android¶ Note You will need Android NDK r15 or higher to build Ceres solver. To build Ceres for Android, we need to force CMake to find the toolchains from the Android NDK instead of using the standard ones. For example, assuming you have specified $NDK_DIR:

cmake \
-DCMAKE_TOOLCHAIN_FILE=\

Specify Ceres components¶

You can specify particular Ceres components that you require (in order for Ceres to be reported as found) when invoking find_package(Ceres). This allows you to specify, for example, that you require a version of Ceres built with SuiteSparse support. By definition, if you do not specify any components when calling find_package(Ceres) (the default) any version of Ceres detected will be reported as found, irrespective of which components it was built with.

The Ceres components which can be specified are:

1. LAPACK: Ceres built using LAPACK (LAPACK=ON).

2. SuiteSparse: Ceres built with SuiteSparse (SUITESPARSE=ON).

3. CXSparse: Ceres built with CXSparse (CXSPARSE=ON).

4. AccelerateSparse: Ceres built with Apple’s Accelerate sparse solvers (ACCELERATESPARSE=ON).

5. EigenSparse: Ceres built with Eigen’s sparse Cholesky factorization (EIGENSPARSE=ON).

6. SparseLinearAlgebraLibrary: Ceres built with at least one sparse linear algebra library. This is equivalent to SuiteSparse OR CXSparse OR AccelerateSparse OR EigenSparse.

7. SchurSpecializations: Ceres built with Schur specializations (SCHUR_SPECIALIZATIONS=ON).

8. OpenMP: Ceres built with OpenMP (CERES_THREADING_MODEL=OPENMP).

9. Multithreading: Ceres built with a multithreading library. This is equivalent to (CERES_THREAD != NO_THREADS).

To specify one/multiple Ceres components use the COMPONENTS argument to find_package() like so:

# Find a version of Ceres compiled with SuiteSparse & EigenSparse support.
#
# NOTE: This will report Ceres as **not** found if the detected version of
#            Ceres was not compiled with both SuiteSparse & EigenSparse.
#            Remember, if you have multiple versions of Ceres installed, you
#            can use Ceres_DIR to specify which should be used.
find_package(Ceres REQUIRED COMPONENTS SuiteSparse EigenSparse)


Specify Ceres version¶

Additionally, when CMake has found Ceres it can optionally check the package version, if it has been specified in the find_package() call. For example:

find_package(Ceres 1.2.3 REQUIRED)


Local installations¶

If Ceres was installed in a non-standard path by specifying -DCMAKE_INSTALL_PREFIX="/some/where/local", then the user should add the PATHS option to the find_package() command, e.g.,

find_package(Ceres REQUIRED PATHS "/some/where/local/")


Note that this can be used to have multiple versions of Ceres installed. However, particularly if you have only a single version of Ceres which you want to use but do not wish to install to a system location, you should consider exporting Ceres using the EXPORT_BUILD_DIR option instead of a local install, as exported versions of Ceres will be automatically detected by CMake, irrespective of their location.

Understanding the CMake Package System¶

Although a full tutorial on CMake is outside the scope of this guide, here we cover some of the most common CMake misunderstandings that crop up when using Ceres. For more detailed CMake usage, the following references are very useful:

• Provides a tour of the core features of CMake.

• Cover how to write a ProjectConfig.cmake file, discussed below, for your own project when installing or exporting it using CMake. It also covers how these processes in conjunction with find_package() are actually handled by CMake. The ProjectConfig tutorial is the older style, currently used by Ceres for compatibility with older versions of CMake.

Note

Targets in CMake.

All libraries and executables built using CMake are represented as targets created using add_library() and add_executable(). Targets encapsulate the rules and dependencies (which can be other targets) required to build or link against an object. This allows CMake to implicitly manage dependency chains. Thus it is sufficient to tell CMake that a library target: B depends on a previously declared library target A, and CMake will understand that this means that B also depends on all of the public dependencies of A.

When a project like Ceres is installed using CMake, or its build directory is exported into the local CMake package registry (see Installing a project with CMake vs Exporting its build directory), in addition to the public headers and compiled libraries, a set of CMake-specific project configuration files are also installed to: <INSTALL_ROOT>/lib/cmake/Ceres (if Ceres is installed), or created in the build directory (if Ceres’ build directory is exported). When find_package is invoked, CMake checks various standard install locations (including /usr/local on Linux & UNIX systems), and the local CMake package registry for CMake configuration files for the project to be found (i.e. Ceres in the case of find_package(Ceres)). Specifically it looks for:

• <PROJECT_NAME>Config.cmake (or <lower_case_project_name>-config.cmake)

Which is written by the developers of the project, and is configured with the selected options and installed locations when the project is built and imports the project targets and/or defines the legacy CMake variables: <PROJECT_NAME>_INCLUDE_DIRS & <PROJECT_NAME>_LIBRARIES which are used by the caller.

The <PROJECT_NAME>Config.cmake typically includes a second file installed to the same location:

• <PROJECT_NAME>Targets.cmake

Which is autogenerated by CMake as part of the install process and defines imported targets for the project in the caller’s CMake scope.

An imported target contains the same information about a library as a CMake target that was declared locally in the current CMake project using add_library(). However, imported targets refer to objects that have already been built by a different CMake project. Principally, an imported target contains the location of the compiled object and all of its public dependencies required to link against it as well as all required include directories. Any locally declared target can depend on an imported target, and CMake will manage the dependency chain, just as if the imported target had been declared locally by the current project.

Crucially, just like any locally declared CMake target, an imported target is identified by its name when adding it as a dependency to another target.

Since v2.0, Ceres has used the target namespace feature of CMake to prefix its export targets: Ceres::ceres. However, historically the Ceres target did not have a namespace, and was just called ceres.

Whilst an alias target called ceres is still provided in v2.0 for backwards compatibility, it creates a potential drawback, if you failed to call find_package(Ceres), and Ceres is installed in a default search path for your compiler, then instead of matching the imported Ceres target, it will instead match the installed libceres.so/dylib/a library. If this happens you will get either compiler errors for missing include directories or linker errors due to missing references to Ceres public dependencies.

Note that this description applies both to projects that are installed using CMake, and to those whose build directory is exported using export() (instead of install()). Ceres supports both installation and export of its build directory if the EXPORT_BUILD_DIR option is enabled, see Customizing the build.

Installing a project with CMake vs Exporting its build directory¶

When a project is installed, the compiled libraries and headers are copied from the source & build directory to the install location, and it is these copied files that are used by any client code. When a project’s build directory is exported, instead of copying the compiled libraries and headers, CMake creates an entry for the project in the user’s local CMake package registry, <USER_HOME>/.cmake/packages on Linux & macOS, which contains the path to the project’s build directory which will be checked by CMake during a call to find_package(). The effect of which is that any client code uses the compiled libraries and headers in the build directory directly, thus not requiring the project to be installed to be used.

Installing / Exporting a project that uses Ceres¶

As described in Understanding the CMake Package System, the contents of the CERES_LIBRARIES variable is the name of an imported target which represents Ceres. If you are installing / exporting your own project which uses Ceres, it is important to understand that:

Imported targets are not (re)exported when a project which imported them is exported.

Thus, when a project Foo which uses Ceres is exported, its list of dependencies as seen by another project Bar which imports Foo via: find_package(Foo REQUIRED) will contain: ceres. However, the definition of ceres as an imported target is not (re)exported when Foo is exported. Hence, without any additional steps, when processing Bar, ceres will not be defined as an imported target. Thus, when processing Bar, CMake will assume that ceres refers only to: libceres.a/so/dylib/lib (the compiled Ceres library) directly if it is on the current list of search paths. In which case, no CMake errors will occur, but Bar will not link properly, as it does not have the required public link dependencies of Ceres, which are stored in the imported target definition.

The solution to this is for Foo (i.e., the project that uses Ceres) to invoke find_package(Ceres) in FooConfig.cmake, thus ceres will be defined as an imported target when CMake processes Bar. An example of the required modifications to FooConfig.cmake are show below:

# Importing Ceres in FooConfig.cmake using CMake 3.x style.
#
# In CMake v3.x, the find_dependency() macro exists to forward the REQUIRED
# / QUIET parameters to find_package() when searching for dependencies.
#
# Note that find_dependency() does not take a path hint, so if Ceres was
# installed in a non-standard location, that location must be added to
# CMake's search list before this call.
include(CMakeFindDependencyMacro)
find_dependency(Ceres)


Migration¶

The following includes some hints for migrating from previous versions.

Version 2.0¶

• When using Ceres with CMake, the target name in v2.0 is Ceres::ceres following modern naming convetions. The legacy target ceres exists for backwards compatibility, but is deprecated. CERES_INCLUDE_DIRS is not set any more, as the exported Ceres CMake target already contains the definitions of its public include directories which will be automatically included by CMake when compiling a target that links against Ceres.

• When building Ceres, some dependencies (Eigen, gflags) are not found using custom Find<DEPENDENCY_NAME>.cmake modules any more. Hence, instead of the custom variables (<DEPENDENCY_NAME (CAPS)>_INCLUDE_DIR_HINTS, <DEPENDENCY_NAME (CAPS)>_INCLUDE_DIR, …) you should use standard CMake facilities to customize where these dependencies are found, such as CMAKE_PREFIX_PATH, the <DEPENDENCY_NAME>_DIR variables, or since CMake 3.12 the <DEPENDENCY_NAME>_ROOT variables.

• While TBB is not used any more directly by Ceres, it might still try to link against it, if SuiteSparseQR was found. The variable (environment or CMake) to customize this is TBB_ROOT (used to be TBBROOT). For example, use cmake -DTBB_ROOT=/opt/intel/tbb ...` if you want to link against TBB installed from Intel’s binary packages on Linux.