Dependencies

WOLF has the following dependencies:

• Mandatory: CMake, Eigen, yaml-cpp, yaml-schema-cpp and spdlog.

• Optional, recommended: SuiteSparse and Ceres.

• Optional: Boost.

The optional dependencies are detected automatically when WOLF is configured with CMake, and the CMake output reports which ones were found.

CMake

Building tool used by WOLF and by some of its dependencies. In order to install cmake please follow the instructions at cmake site. WOLF requires >= CMake-3.18 which can usually be installed via repository.

Eigen

Eigen is a linear algebra, header library. WOLF requires Eigen >= 3.3.7. We recommend to install it via apt.

sudo apt install libeigen3-dev

yaml-cpp

WOLF uses YAML files for configuration and for saving and loading workspaces. Minimum version required is 0.7.

From ubuntu 22.04, it can be installed via apt:

sudo apt install libyaml-cpp-dev

In previous ubuntu distributions, it has to be installed via source.

yaml-schema-cpp

yaml-schema-cpp is a library for specifying and checking the structure, entries and types of YAML files.

To install it:

git clone -b main --depth 1 https://github.com/joanvallve/yaml-schema-cpp.git
cd yaml-schema-cpp
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=release -DBUILD_EXAMPLES=ON -DBUILD_TESTS=ON ..
make
sudo make install

Note

To install it in a local folder without sudo, see local installation.

spdlog

WOLF uses spdlog library for logging. Minimum version required is v1.0.0. It can be installed via apt:

sudo apt install libspdlog-dev

SuiteSparse (optional, recommended)

SuiteSparse is a collection of sparse matrix libraries. WOLF uses two of them in SolverEigen, the solver of WOLF core that does not depend on Ceres (see next section):

• CHOLMOD (supernodal sparse Cholesky) for linear_solver: cholesky.

• SPQR (multifrontal sparse QR) for linear_solver: qr.

Without SuiteSparse, SolverEigen uses Eigen’s built-in sparse solvers (SimplicialLDLT and SparseQR). They give the same results but are much slower for large problems, especially the QR, which can be orders of magnitude slower. Ceres also uses SuiteSparse if it is installed before compiling Ceres.

It can be installed via apt:

sudo apt install libsuitesparse-dev

When configuring WOLF, CMake reports whether it was found:

-- SuiteSparse found: SolverEigen will use SPQR and CHOLMOD

Note

Both SuiteSparse 7 or newer (which provides CMake configuration files) and older versions (e.g. the one of ubuntu 22.04) are supported.

Ceres (optional, recommended)

Ceres is an optimization library. If it is found, WOLF core includes the solver SolverCeres. Otherwise, SolverEigen is the only solver available.

We recommend installing it:

• SolverCeres provides trust-region methods (e.g. Levenberg-Marquardt) and many more options than SolverEigen. It can be selected in any configuration file (type: SolverCeres).

• The tests of SolverCeres and its comparison with SolverEigen in demo_solvers_comparison are only built if Ceres was found.

All Ceres dependencies can be installed via apt. They include SuiteSparse, which Ceres uses if it is installed before compiling Ceres:

sudo apt install libgoogle-glog-dev libgflags-dev libatlas-base-dev libsuitesparse-dev

Ceres 2.2 can be installed:

wget http://ceres-solver.org/ceres-solver-2.2.0.tar.gz
tar zxf ceres-solver-2.2.0.tar.gz
rm ceres-solver-2.2.0.tar.gz
mkdir ceres-bin
cd ceres-bin
cmake ../ceres-solver-2.2.0
make
make test
sudo make install

When configuring WOLF, CMake reports whether it was found:

-- Ceres 2.2.0 found: SolverCeres will be built
-- Ceres not found: SolverCeres will not be built (only SolverEigen)

To build WOLF without Ceres even if it is installed, add -DCMAKE_DISABLE_FIND_PACKAGE_Ceres=ON to the cmake command.

Boost (optional)

Boost 1.73 or higher is an optional dependency. If found, the macro Macros for processors is fully operative.