We compute the minimizer of the convex function

whereby

• E are the edges of a tree graph,
• λ > 0 some edge weights,
• μ ≥ 0 some node weights.

From left to right:

1. Original Image
2. Gaussian noise added (std=0.25)
3. E is grid graph (λ = 0.2)
4. E is random spanning tree (λ = 0.2)

... is realized via pybind11. For compiling, add the Git submodule

git submodule update --init --recursive deps/graphidx

Afterwards build the python extension by calling setup.py

cd python && python setup.py develop

├── ci          Continous integration
├── cxx         C++ source/header files
├── data        Some test data and benchmarks
├── deps        External dependencies/libraries
├── julia       Julia Bindings/Implementation
├── rust        Rust re-implementation
└── python      Python bindings

There are several implementations to be able assess e.g. different memory layouts, traversing orders, etc.

We also provide specialized algorithms for one dimensional line graphs (also called “trail graphs”). For the sake of benchmarks the implementation by Laurent Condat and the glmgen R package are included as well.