Create automatic benchmarks for comparing methods.

[FHorschig]

As Dr. Sack suggested, we should find an easy (automated) way to compare the results of our algorithms against a selection of images with predefined results.

--> This allows to evaluate if changes improved the algorithm for more than one picture without having to start it for several images.

Focus: Count of found objects. Bounding boxes overlap expected ones.