REFACTOR PLAN

[Hendrik-code]

Refactor ideas in order:

• Need a module between end of instance evaluation and Panoptica Result, that actually handles metric computation -> PanopticaResult object becomes a purely result-keeping object.

• refactor result objects for each metric type (global, instance, region-wise, autc), to be easily expandable and still understandable by the user (something like result.global.dice vs. result.autc.dice)

• refactor whole codebase in terms of folder/file structure, make folders for each pipeline phase, sort utils better, ...

• Large refactor of the whole pipeline, switching to a metric-focused approach (e.g. we loop over each metric the user wants, compute it, save all intermediate steps so that other metrics can still utilize the intermediate steps instead of computing everything anew).

• Speed improvements?

[neuronflow]

I like the idea of having three result objects:

• global: voxel-/pixel-wise
• voronoi
• instance-wise