Embedding benchmarking and quantification capability into Savu software

Savu is a Python package, developed in the Data Analysis Group, to assist with the processing and reconstruction of parallel-beam tomography data. There has been a lot of recent interest in Savu in the greater tomography community, both on-site (more Diamond beamlines, IMAT@ISIS and CLF), at Universities and internationally at other synchrotrons (MAX IV, PETRA III, NSLS II) and the range of experiments that Savu can process is subsequently growing. Its popularity is gained from its flexibility, with collaborators encouraged to develop their own methods or integrate existing ones, in the form of Savu plugins.

As the number of plugins (and methods) increases in Savu, it is important to ensure the integrity of these methods to produce robust scientific output. It is also possible to have multiple implementations of the same method, where the implementation of choice is often based on familiarity rather than scientific rigour. All too often there is no mention of the accuracy of a processing method in a scientific paper, which is critical to ensuring the authenticity of the result.
The Savu modular plugin design can be utilised to add “modelling” and “quantification” plugins to flank a chosen list of processing plugin(s), such as a new method in development or a standard processing pipeline already used on the tomography beamlines. The “modelling” plugins (already available in Savu to some extent) would create realistic simulated data from the known sample (a ground truth). The “quantification” plugins will be used to quantify each processing chain, in terms of its accuracy, by calculating some statistical measures between the ground truth and the obtained output.

This project is a logical continuation of already existing developments of the group and will ensure a smooth transaction to new elements during its progress.