1 oracle.pipelines.evaluated_plotting

oracle.pipelines.evaluated_plotting

Canonical figures for in-memory Oracle-evaluated rollout evidence.

The module turns aria_nbv.oracle.pipelines.evaluated_inspection products into immutable Plotly JSON plus thesis-aligned scientific explanations. It does not acquire samples, execute replay, mutate evidence, or import Streamlit. UI adapters may rebuild a fresh mutable Plotly figure for view-specific controls without recomputing scientific reductions.

1.1 Classes

Name Description
EvaluatedPlot Immutable Plotly payload paired with its scientific interpretation.
EvaluatedSceneFigureOptions Resolved display-only options for a live rollout 3D evidence view.

1.1.1 EvaluatedPlot

oracle.pipelines.evaluated_plotting.EvaluatedPlot(key, plotly_json, explanation)

Immutable Plotly payload paired with its scientific interpretation.

1.1.1.1 Attributes

Name Description
key Stable result-local figure identifier.
plotly_json Canonical UTF-8 Plotly JSON with generated trace UIDs removed.
explanation Domain-owned narrative and canonical notation references.

1.1.1.2 Methods

Name Description
build_figure Build a fresh mutable Plotly figure for one presentation render.
1.1.1.2.1 build_figure
oracle.pipelines.evaluated_plotting.EvaluatedPlot.build_figure()

Build a fresh mutable Plotly figure for one presentation render.

1.1.2 EvaluatedSceneFigureOptions

oracle.pipelines.evaluated_plotting.EvaluatedSceneFigureOptions(
    camera='rgb',
    show_mesh=True,
    mesh_opacity=0.18,
    show_source_trajectory=False,
    mark_source_trajectory_endpoints=True,
    show_scene_bounds=False,
    show_crop_bounds=False,
    show_source_frustum=False,
    source_frustum_frame_indices=(0,),
    source_frustum_scale=0.7,
    show_all_gt_obbs=False,
    gt_timestamp=None,
    semidense_mode='off',
    max_semidense_points=20000,
    show_actor_target=True,
    show_gt_target=True,
    show_target_semidense_crop=False,
    target_crop_basis='descriptor',
    show_selected_frusta=True,
    selected_frustum_scale=0.45,
    include_rejected_candidates=False,
    show_candidate_frusta=True,
    candidate_frustum_scale=0.5,
    max_candidate_frusta=16,
    candidate_color_metric='selection_score',
)

Resolved display-only options for a live rollout 3D evidence view.

1.1.2.1 Attributes

Name Description
camera Source-camera stream used for snippet frusta and optional GT overlays.
show_mesh Whether to show the privileged ASE scene mesh.
mesh_opacity Mesh opacity in the interval (0, 1].
show_source_trajectory Whether to show the observed device trajectory.
mark_source_trajectory_endpoints Whether the observed trajectory marks its first and last poses.
show_scene_bounds Whether to show world-frame scene bounds.
show_crop_bounds Whether to show configured mesh-crop bounds.
show_source_frustum Whether to show selected source-camera frusta.
source_frustum_frame_indices Source frame indices whose frusta are displayed.
source_frustum_scale Positive display scale for source-camera frusta.
show_all_gt_obbs Whether to show all privileged GT OBB annotations.
gt_timestamp Annotation timestamp used for the optional GT population overlay.
semidense_mode Observed semidense point display mode.
max_semidense_points Positive display-only cap on observed semidense points.
show_actor_target Whether to show the actor-visible detected target descriptor.
show_gt_target Whether to show the uniquely matched privileged GT target.
show_target_semidense_crop Whether to show semidense points inside the selected target OBB.
target_crop_basis Actor descriptor or privileged GT OBB used for the display-only crop.
show_selected_frusta Whether path views show each selected camera frustum.
selected_frustum_scale Positive display scale for selected camera frusta.
include_rejected_candidates Whether step-shell views include hard-invalid candidate centres.
show_candidate_frusta Whether step-shell views show valid candidate frusta.
candidate_frustum_scale Positive display scale for candidate frusta.
max_candidate_frusta Optional display-only candidate-frustum cap; never a statistical filter.
candidate_color_metric Actor-side value used to colour valid candidate rows.

1.2 Functions

Name Description
build_candidate_figures Build candidate allocation, rejection, and policy-score figures for one step.
build_reward_figures Build reward, candidate-range, headroom, and endpoint figures.
build_rollout_scene_figure Build one complete 3D path or candidate-shell evidence figure.
build_selected_depth_figure Build a metric selected-depth figure from already retained evidence.
build_selected_target_depth_overlays Build actor-descriptor and matched-GT overlays for selected depth.
project_selected_depth_box Project one world-frame OBB corner set into retained selected-depth pixels.

1.2.1 build_candidate_figures

oracle.pipelines.evaluated_plotting.build_candidate_figures(step)

Build candidate allocation, rejection, and policy-score figures for one step.

1.2.2 build_reward_figures

oracle.pipelines.evaluated_plotting.build_reward_figures(inspection)

Build reward, candidate-range, headroom, and endpoint figures.

Candidate ribbons show the within-step empirical 2.5th–97.5th percentile range of hard-valid candidates. They are neither a confidence interval nor a between-trajectory IQR. Selected and cumulative trajectories preserve observed zeros, negative rewards, and absent early-terminated tails.

1.2.3 build_rollout_scene_figure

oracle.pipelines.evaluated_plotting.build_rollout_scene_figure(
    sample,
    evaluated,
    *,
    target,
    view='paths',
    trajectory_index=0,
    step_index=0,
    options=None,
)

Build one complete 3D path or candidate-shell evidence figure.

The sample must already contain its live EFM snippet. This function only converts existing in-memory evidence to Plotly traces; it performs no store read, target admission, candidate generation, rendering, or scoring.

1.2.4 build_selected_depth_figure

oracle.pipelines.evaluated_plotting.build_selected_depth_figure(
    depth,
    *,
    overlays=(),
)

Build a metric selected-depth figure from already retained evidence.

1.2.5 build_selected_target_depth_overlays

oracle.pipelines.evaluated_plotting.build_selected_target_depth_overlays(
    depth,
    sample,
    target,
    *,
    show_actor_target,
    show_gt_target,
)

Build actor-descriptor and matched-GT overlays for selected depth.

Actor-visible descriptor geometry and privileged GT evaluation geometry retain distinct names and colours. Unrequested or unavailable target evidence produces no overlay rather than an inferred box.

1.2.6 project_selected_depth_box

oracle.pipelines.evaluated_plotting.project_selected_depth_box(
    depth,
    corners_world,
    *,
    name,
    color,
    width=4,
)

Project one world-frame OBB corner set into retained selected-depth pixels.

1.2.6.1 Parameters

Name Type Description Default
depth SelectedDepthEvidence Available selected-depth evidence with retained intrinsics and selected world-from-camera pose. required
corners_world Tensor["8 3", float32] OBB corners in world-frame metres. required
name str Legend label identifying the evidence role of the box. required
color str Plotly-compatible line colour. required
width int Positive overlay line width in pixels. 4
Back to top