1 OracleRRI

rri_metrics.oracle_rri.OracleRRI(config)

Facade to compute oracle RRI for one or more candidates.

1.1 Conceptual steps

Merge P_t (current eval points) with candidate view point cloud P_q to obtain P_{t∪q}.
(Optional) Voxel-downsample both P_t and P_{t∪q} to ensure comparable density when evaluating point-mesh distances.
Compute accuracy/completeness distances to the GT mesh using the PyTorch3D backend.
Form RRI = (d_before - d_after) / d_before and return diagnostics.

1.2 Methods

Name	Description
score	Compute RRI for one or more candidates in a single forward pass.
score_batch	Alias kept for callers using the old batch name.

1.2.1 score

rri_metrics.oracle_rri.OracleRRI.score(
    points_t,
    points_q,
    lengths_q,
    gt_verts,
    gt_faces,
    extend,
)

Compute RRI for one or more candidates in a single forward pass.

1.2.1.1 Parameters

Name	Type	Description	Default
points_t	torch.Tensor	`Tensor['N_t', 3]` current eval point cloud up to time t.	required
points_q	torch.Tensor	`Tensor['N_q', 3]` candidate-view point cloud rendered from GT.	required
gt_verts	torch.Tensor	`Tensor['V', 3]` ground-truth mesh vertices.	required
gt_faces	torch.Tensor	`Tensor['F', 3]` ground-truth mesh face indices (int64).	required
extend	torch.Tensor	`Tensor[6]` [xmin, xmax, ymin, ymax, zmin, zmax] AABB in world frame used to crop the GT mesh.	required

Returns: RriResult containing scalar RRI and distance breakdowns.

1.2.2 score_batch

rri_metrics.oracle_rri.OracleRRI.score_batch(
    points_t,
    points_q,
    lengths_q,
    gt_verts,
    gt_faces,
    extend,
)

Alias kept for callers using the old batch name.

# OracleRRI { #aria_nbv.rri_metrics.oracle_rri.OracleRRI } ```python rri_metrics.oracle_rri.OracleRRI(config) ``` Facade to compute oracle RRI for one or more candidates. ## Conceptual steps {.doc-section .doc-section-conceptual-steps} 1. Merge ``P_t`` (current eval points) with candidate view point cloud ``P_q`` to obtain ``P_{t∪q}``. 2. (Optional) Voxel-downsample both ``P_t`` and ``P_{t∪q}`` to ensure comparable density when evaluating point-mesh distances. 3. Compute accuracy/completeness distances to the GT mesh using the PyTorch3D backend. 4. Form RRI = (d_before - d_after) / d_before and return diagnostics. ## Methods | Name | Description | | --- | --- | | [score](#aria_nbv.rri_metrics.oracle_rri.OracleRRI.score) | Compute RRI for one or more candidates in a single forward pass. | | [score_batch](#aria_nbv.rri_metrics.oracle_rri.OracleRRI.score_batch) | Alias kept for callers using the old batch name. | ### score { #aria_nbv.rri_metrics.oracle_rri.OracleRRI.score } ```python rri_metrics.oracle_rri.OracleRRI.score( points_t, points_q, lengths_q, gt_verts, gt_faces, extend, ) ``` Compute RRI for one or more candidates in a single forward pass. #### Parameters {.doc-section .doc-section-parameters} | Name | Type | Description | Default | |----------|--------------|--------------------------------------------------------------------------------------------------|------------| | points_t | torch.Tensor | ``Tensor['N_t', 3]`` current eval point cloud up to time *t*. | _required_ | | points_q | torch.Tensor | ``Tensor['N_q', 3]`` candidate-view point cloud rendered from GT. | _required_ | | gt_verts | torch.Tensor | ``Tensor['V', 3]`` ground-truth mesh vertices. | _required_ | | gt_faces | torch.Tensor | ``Tensor['F', 3]`` ground-truth mesh face indices (int64). | _required_ | | extend | torch.Tensor | ``Tensor[6]`` [xmin, xmax, ymin, ymax, zmin, zmax] AABB in world frame used to crop the GT mesh. | _required_ | Returns: ``RriResult`` containing scalar RRI and distance breakdowns. ### score_batch { #aria_nbv.rri_metrics.oracle_rri.OracleRRI.score_batch } ```python rri_metrics.oracle_rri.OracleRRI.score_batch( points_t, points_q, lengths_q, gt_verts, gt_faces, extend, ) ``` Alias kept for callers using the old batch name.