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 current tensor APIs use a rerun-specific tensor format that stores tensor-dimension as a data field, giving us support for dynamic-shape-tensor.
Using fixed-shape-tensor ourselves would be quite a bit of work at this point with two major challenges:
We need a mechanism to surface metadata from components.
We need to be careful about how metadata is tracked and impacted by things like batching.
That said, users that are already working with arrow tensors from some other library would benefit from ergonomic support for forwarding these on into Rerun directly. This might be another interesting use-case for derived components.
The text was updated successfully, but these errors were encountered:
The arrow project has an official extension for fixed-shape-tensors: https://arrow.apache.org/docs/dev/format/CanonicalExtensions.html#fixed-shape-tensor
Our current tensor APIs use a rerun-specific tensor format that stores tensor-dimension as a data field, giving us support for dynamic-shape-tensor.
Using fixed-shape-tensor ourselves would be quite a bit of work at this point with two major challenges:
That said, users that are already working with arrow tensors from some other library would benefit from ergonomic support for forwarding these on into Rerun directly. This might be another interesting use-case for derived components.
The text was updated successfully, but these errors were encountered: