- GUI or napari plugin
- limit output to informative messages
- informative error message when registration between pair of views fails, and possibly save input images to temp dir
- add ipython notebook for debugging (napari?)
###Performance
- perform blockwise transformations to handle large target volumes
- optimize weight calculation (downscaling seems to take long) (dask.array.coarsen?)
- be tolerant to wrong registrations
- build graph for registrations and resolve shortest paths
- build candidate pairs from metadata (e.g. group in y, define weights by difference of angle, distance between middle of stacks)