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
Newly-default spatial tree library kiddo supports zero-copy serialisation/ deserialisation with rkyv, which means that the structs which are practically thin wrappers around it (i.e. this whole crate) should be able to as well without too much hassle. This would allow for larger-than-RAM arenas, or at least extremely efficient caching. Currently CATMAID uses .rda objects to cache resampled neurons between runs; this should be much better.
Currently, kiddo's neuron objects are much bigger than necessary because the points are duplicated, but that will not be necessary once sdd/kiddo#135 is merged.
The text was updated successfully, but these errors were encountered:
At present, just recreating the tree from points, tangents, alphas is much faster than deserialising from CBOR, at least [1]. That isn't zero-copy, but it's not the copying which is the slow bit here.
Newly-default spatial tree library kiddo supports zero-copy serialisation/ deserialisation with rkyv, which means that the structs which are practically thin wrappers around it (i.e. this whole crate) should be able to as well without too much hassle. This would allow for larger-than-RAM arenas, or at least extremely efficient caching. Currently CATMAID uses .rda objects to cache resampled neurons between runs; this should be much better.
Currently, kiddo's neuron objects are much bigger than necessary because the points are duplicated, but that will not be necessary once sdd/kiddo#135 is merged.
The text was updated successfully, but these errors were encountered: