You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Our choice of axis ordering is reasonable when one wants to deal with projection images one at a time, but for other purposes it may be more reasonable to have a different ordering (for example to match the ordering of the backend, thereby removing overhead in copying and transposing data).
My suggestion for geometries in general is that we remove the lumped together attributes partition, params and grid, thus making geometries agnostic of axis ordering. The "consumers" of geometries (i.e. mostly RayTransform) should be responsible for handling axes in the data.
I think this change would only affect internal stuff and not break any user code.
The text was updated successfully, but these errors were encountered:
Frankly I agree. The current ordering is somewhat arbitrary to enforce on a global scale. It also causes issues for geometries that don't really have any motion.
Our choice of axis ordering is reasonable when one wants to deal with projection images one at a time, but for other purposes it may be more reasonable to have a different ordering (for example to match the ordering of the backend, thereby removing overhead in copying and transposing data).
My suggestion for geometries in general is that we remove the lumped together attributes
partition
,params
andgrid
, thus making geometries agnostic of axis ordering. The "consumers" of geometries (i.e. mostlyRayTransform
) should be responsible for handling axes in the data.I think this change would only affect internal stuff and not break any user code.
The text was updated successfully, but these errors were encountered: