1 vin.diagnostics.plotting

vin.diagnostics.plotting

VIN plotting helpers for geometry, feature, and prediction diagnostics.

This module provides Plotly figures for candidate poses, voxel evidence, semidense projections, learned encodings, and frame-consistency checks.

1.1 Functions

Name	Description
build_backbone_evidence_figures	Plot sparse voxel evidence as world-space scatter plots.
build_field_slice_figures	Plot XY/XZ/YZ orthogonal slices for selected channels.
build_geometry_overview_figure	Plot grid bounds, reference/voxel axes, and candidate centers in one figure.
build_lff_empirical_figures	Empirical distributions for LFF Fourier + MLP outputs using actual pose_vec.
build_pos_grid_linearity_figure	Check linearity between voxel coordinates and pos_grid values.
build_pose_enc_pca_figure	Project flattened pose encodings to a two-dimensional PCA figure.
build_pose_grid_pca_figure	Project positional grid embeddings to 2D via PCA.
build_pose_grid_slices_figure	Plot 2D slices of the normalized position grid (x/y/z channels).
build_pose_vec_histogram	Histogram of a single pose vector component.
build_scene_field_evidence_figures	Plot scene-field channels (field_in) as world-space scatter plots.
build_se3_closure_figure	Check closure consistency of T_world_cam = T_world_rig_ref * T_rig_ref_cam.
build_semidense_cnn_grid_figure	Visualize the projection grid maps used by the VIN v3 semidense CNN.
build_semidense_cnn_grid_maps	Compute semidense grid maps that match VIN v3’s tiny-CNN inputs.
build_semidense_projection_feature_figure	Visualize semidense projection feature maps for a candidate.
build_semidense_projection_feature_maps	Compute semidense projection feature maps for a candidate view.
build_semidense_projection_figure	Plot semidense points colored by whether they project inside the candidate view.
build_valid_fraction_figure	Plot candidate centers colored by validity fraction.
build_voxel_frame_figure	Plot voxel grid bounds and voxel axes in world coordinates.
build_voxel_inbounds_figure	Plot candidate in-bounds ratios and normalized coordinate distributions.
build_voxel_roundtrip_figure	Visualize metric residuals from a voxel-to-world-to-voxel round trip.

1.1.1 build_backbone_evidence_figures

vin.diagnostics.plotting.build_backbone_evidence_figures(
    debug,
    *,
    occ_threshold=0.5,
    max_points=40000,
)

Plot sparse voxel evidence as world-space scatter plots.

1.1.2 build_field_slice_figures

vin.diagnostics.plotting.build_field_slice_figures(
    field,
    *,
    channel_names,
    max_channels=4,
    slice_indices=None,
    title_prefix='field',
)

Plot XY/XZ/YZ orthogonal slices for selected channels.

1.1.3 build_geometry_overview_figure

vin.diagnostics.plotting.build_geometry_overview_figure(
    debug,
    *,
    snippet,
    reference_pose_world_rig,
    max_candidates=64,
    show_reference_axes=True,
    show_voxel_axes=True,
    display_rotate_yaw_cw90=False,
    show_scene_bounds=True,
    show_crop_bounds=False,
    show_frustum=False,
    frustum_camera='rgb',
    frustum_frame_indices=None,
    frustum_scale=1.0,
    show_gt_obbs=False,
    gt_timestamp=None,
    semidense_mode='off',
    max_sem_points=20000,
    show_trajectory=False,
    mark_first_last=True,
    candidate_pose_mode='ref_rig',
    candidate_poses_world_cam=None,
    candidate_color_mode='valid_fraction',
    candidate_color='#ffd966',
    candidate_colorscale='Viridis',
    candidate_color_values=None,
    candidate_color_title='value',
    candidate_frusta_indices=None,
    candidate_frusta_camera='rgb',
    candidate_frusta_frame_index=None,
    candidate_frusta_scale=0.5,
    candidate_frusta_color='#ff4d4d',
    candidate_frusta_show_axes=False,
    candidate_frusta_show_center=False,
    backbone_fields=None,
    backbone_occ_threshold=0.5,
    backbone_max_points=40000,
    backbone_colorscale='Viridis',
)

Plot grid bounds, reference/voxel axes, and candidate centers in one figure.

1.1.4 build_lff_empirical_figures

vin.diagnostics.plotting.build_lff_empirical_figures(
    pose_vec,
    pose_encoder,
    *,
    max_features=96,
    hist_bins=60,
    max_points=8000,
    log1p_counts=False,
)

Empirical distributions for LFF Fourier + MLP outputs using actual pose_vec.

1.1.5 build_pos_grid_linearity_figure

vin.diagnostics.plotting.build_pos_grid_linearity_figure(
    pos_grid,
    voxel_extent,
    *,
    max_points=50000,
)

Check linearity between voxel coordinates and pos_grid values.

1.1.6 build_pose_enc_pca_figure

vin.diagnostics.plotting.build_pose_enc_pca_figure(
    pose_enc,
    *,
    color_values=None,
    color_label='value',
    max_points=8000,
)

Project flattened pose encodings to a two-dimensional PCA figure.

1.1.7 build_pose_grid_pca_figure

vin.diagnostics.plotting.build_pose_grid_pca_figure(
    pos_grid,
    *,
    pos_proj,
    max_points=8000,
    color_by='radius',
    show_axes=True,
    axis_scale=0.5,
)

Project positional grid embeddings to 2D via PCA.

1.1.8 build_pose_grid_slices_figure

vin.diagnostics.plotting.build_pose_grid_slices_figure(pos_grid, *, axis, index)

Plot 2D slices of the normalized position grid (x/y/z channels).

1.1.9 build_pose_vec_histogram

vin.diagnostics.plotting.build_pose_vec_histogram(
    pose_vec,
    *,
    dim_index,
    num_bins=60,
    log1p_counts=False,
)

Histogram of a single pose vector component.

1.1.10 build_scene_field_evidence_figures

vin.diagnostics.plotting.build_scene_field_evidence_figures(
    debug,
    *,
    channel_names,
    occ_threshold=0.5,
    max_points=40000,
)

Plot scene-field channels (field_in) as world-space scatter plots.

1.1.11 build_se3_closure_figure

vin.diagnostics.plotting.build_se3_closure_figure(
    candidate_poses_world_cam,
    reference_pose_world_rig,
    *,
    bins=50,
    log1p_counts=False,
)

Check closure consistency of T_world_cam = T_world_rig_ref * T_rig_ref_cam.

1.1.12 build_semidense_cnn_grid_figure

vin.diagnostics.plotting.build_semidense_cnn_grid_figure(
    points_world,
    *,
    p3d_cameras,
    candidate_index,
    grid_size,
    max_points,
)

Visualize the projection grid maps used by the VIN v3 semidense CNN.

1.1.13 build_semidense_cnn_grid_maps

vin.diagnostics.plotting.build_semidense_cnn_grid_maps(
    points_world,
    *,
    p3d_cameras,
    candidate_index,
    grid_size,
    max_points,
)

Compute semidense grid maps that match VIN v3’s tiny-CNN inputs.

1.1.14 build_semidense_projection_feature_figure

vin.diagnostics.plotting.build_semidense_projection_feature_figure(
    points_world,
    *,
    p3d_cameras,
    candidate_index,
    grid_size,
    max_points,
    semidense_obs_count_max,
    semidense_inv_dist_std_min,
    semidense_inv_dist_std_p95,
)

Visualize semidense projection feature maps for a candidate.

1.1.15 build_semidense_projection_feature_maps

vin.diagnostics.plotting.build_semidense_projection_feature_maps(
    points_world,
    *,
    p3d_cameras,
    candidate_index,
    grid_size,
    max_points,
    semidense_obs_count_max,
    semidense_inv_dist_std_min,
    semidense_inv_dist_std_p95,
)

Compute semidense projection feature maps for a candidate view.

1.1.16 build_semidense_projection_figure

vin.diagnostics.plotting.build_semidense_projection_figure(
    points_world,
    *,
    p3d_cameras,
    candidate_index,
    max_points=20000,
    show_frustum=False,
    frustum_scale=None,
    frustum_color='#ff4d4d',
)

Plot semidense points colored by whether they project inside the candidate view.

1.1.17 build_valid_fraction_figure

vin.diagnostics.plotting.build_valid_fraction_figure(debug)

Plot candidate centers colored by validity fraction.

1.1.18 build_voxel_frame_figure

vin.diagnostics.plotting.build_voxel_frame_figure(debug, *, axis_scale=0.5)

Plot voxel grid bounds and voxel axes in world coordinates.

1.1.19 build_voxel_inbounds_figure

vin.diagnostics.plotting.build_voxel_inbounds_figure(
    candidate_poses_world_cam,
    t_world_voxel,
    voxel_extent,
    *,
    bins=50,
)

Plot candidate in-bounds ratios and normalized coordinate distributions.

1.1.20 build_voxel_roundtrip_figure

vin.diagnostics.plotting.build_voxel_roundtrip_figure(
    debug,
    *,
    num_points=2000,
    log1p_counts=False,
)

Visualize metric residuals from a voxel-to-world-to-voxel round trip.

# vin.diagnostics.plotting { #aria_nbv.vin.diagnostics.plotting } `vin.diagnostics.plotting` VIN plotting helpers for geometry, feature, and prediction diagnostics. This module provides Plotly figures for candidate poses, voxel evidence, semidense projections, learned encodings, and frame-consistency checks. ## Functions | Name | Description | | --- | --- | | [build_backbone_evidence_figures](#aria_nbv.vin.diagnostics.plotting.build_backbone_evidence_figures) | Plot sparse voxel evidence as world-space scatter plots. | | [build_field_slice_figures](#aria_nbv.vin.diagnostics.plotting.build_field_slice_figures) | Plot XY/XZ/YZ orthogonal slices for selected channels. | | [build_geometry_overview_figure](#aria_nbv.vin.diagnostics.plotting.build_geometry_overview_figure) | Plot grid bounds, reference/voxel axes, and candidate centers in one figure. | | [build_lff_empirical_figures](#aria_nbv.vin.diagnostics.plotting.build_lff_empirical_figures) | Empirical distributions for LFF Fourier + MLP outputs using actual pose_vec. | | [build_pos_grid_linearity_figure](#aria_nbv.vin.diagnostics.plotting.build_pos_grid_linearity_figure) | Check linearity between voxel coordinates and pos_grid values. | | [build_pose_enc_pca_figure](#aria_nbv.vin.diagnostics.plotting.build_pose_enc_pca_figure) | Project flattened pose encodings to a two-dimensional PCA figure. | | [build_pose_grid_pca_figure](#aria_nbv.vin.diagnostics.plotting.build_pose_grid_pca_figure) | Project positional grid embeddings to 2D via PCA. | | [build_pose_grid_slices_figure](#aria_nbv.vin.diagnostics.plotting.build_pose_grid_slices_figure) | Plot 2D slices of the normalized position grid (x/y/z channels). | | [build_pose_vec_histogram](#aria_nbv.vin.diagnostics.plotting.build_pose_vec_histogram) | Histogram of a single pose vector component. | | [build_scene_field_evidence_figures](#aria_nbv.vin.diagnostics.plotting.build_scene_field_evidence_figures) | Plot scene-field channels (field_in) as world-space scatter plots. | | [build_se3_closure_figure](#aria_nbv.vin.diagnostics.plotting.build_se3_closure_figure) | Check closure consistency of T_world_cam = T_world_rig_ref * T_rig_ref_cam. | | [build_semidense_cnn_grid_figure](#aria_nbv.vin.diagnostics.plotting.build_semidense_cnn_grid_figure) | Visualize the projection grid maps used by the VIN v3 semidense CNN. | | [build_semidense_cnn_grid_maps](#aria_nbv.vin.diagnostics.plotting.build_semidense_cnn_grid_maps) | Compute semidense grid maps that match VIN v3's tiny-CNN inputs. | | [build_semidense_projection_feature_figure](#aria_nbv.vin.diagnostics.plotting.build_semidense_projection_feature_figure) | Visualize semidense projection feature maps for a candidate. | | [build_semidense_projection_feature_maps](#aria_nbv.vin.diagnostics.plotting.build_semidense_projection_feature_maps) | Compute semidense projection feature maps for a candidate view. | | [build_semidense_projection_figure](#aria_nbv.vin.diagnostics.plotting.build_semidense_projection_figure) | Plot semidense points colored by whether they project inside the candidate view. | | [build_valid_fraction_figure](#aria_nbv.vin.diagnostics.plotting.build_valid_fraction_figure) | Plot candidate centers colored by validity fraction. | | [build_voxel_frame_figure](#aria_nbv.vin.diagnostics.plotting.build_voxel_frame_figure) | Plot voxel grid bounds and voxel axes in world coordinates. | | [build_voxel_inbounds_figure](#aria_nbv.vin.diagnostics.plotting.build_voxel_inbounds_figure) | Plot candidate in-bounds ratios and normalized coordinate distributions. | | [build_voxel_roundtrip_figure](#aria_nbv.vin.diagnostics.plotting.build_voxel_roundtrip_figure) | Visualize metric residuals from a voxel-to-world-to-voxel round trip. | ### build_backbone_evidence_figures { #aria_nbv.vin.diagnostics.plotting.build_backbone_evidence_figures } ```python vin.diagnostics.plotting.build_backbone_evidence_figures( debug, *, occ_threshold=0.5, max_points=40000, ) ``` Plot sparse voxel evidence as world-space scatter plots. ### build_field_slice_figures { #aria_nbv.vin.diagnostics.plotting.build_field_slice_figures } ```python vin.diagnostics.plotting.build_field_slice_figures( field, *, channel_names, max_channels=4, slice_indices=None, title_prefix='field', ) ``` Plot XY/XZ/YZ orthogonal slices for selected channels. ### build_geometry_overview_figure { #aria_nbv.vin.diagnostics.plotting.build_geometry_overview_figure } ```python vin.diagnostics.plotting.build_geometry_overview_figure( debug, *, snippet, reference_pose_world_rig, max_candidates=64, show_reference_axes=True, show_voxel_axes=True, display_rotate_yaw_cw90=False, show_scene_bounds=True, show_crop_bounds=False, show_frustum=False, frustum_camera='rgb', frustum_frame_indices=None, frustum_scale=1.0, show_gt_obbs=False, gt_timestamp=None, semidense_mode='off', max_sem_points=20000, show_trajectory=False, mark_first_last=True, candidate_pose_mode='ref_rig', candidate_poses_world_cam=None, candidate_color_mode='valid_fraction', candidate_color='#ffd966', candidate_colorscale='Viridis', candidate_color_values=None, candidate_color_title='value', candidate_frusta_indices=None, candidate_frusta_camera='rgb', candidate_frusta_frame_index=None, candidate_frusta_scale=0.5, candidate_frusta_color='#ff4d4d', candidate_frusta_show_axes=False, candidate_frusta_show_center=False, backbone_fields=None, backbone_occ_threshold=0.5, backbone_max_points=40000, backbone_colorscale='Viridis', ) ``` Plot grid bounds, reference/voxel axes, and candidate centers in one figure. ### build_lff_empirical_figures { #aria_nbv.vin.diagnostics.plotting.build_lff_empirical_figures } ```python vin.diagnostics.plotting.build_lff_empirical_figures( pose_vec, pose_encoder, *, max_features=96, hist_bins=60, max_points=8000, log1p_counts=False, ) ``` Empirical distributions for LFF Fourier + MLP outputs using actual pose_vec. ### build_pos_grid_linearity_figure { #aria_nbv.vin.diagnostics.plotting.build_pos_grid_linearity_figure } ```python vin.diagnostics.plotting.build_pos_grid_linearity_figure( pos_grid, voxel_extent, *, max_points=50000, ) ``` Check linearity between voxel coordinates and pos_grid values. ### build_pose_enc_pca_figure { #aria_nbv.vin.diagnostics.plotting.build_pose_enc_pca_figure } ```python vin.diagnostics.plotting.build_pose_enc_pca_figure( pose_enc, *, color_values=None, color_label='value', max_points=8000, ) ``` Project flattened pose encodings to a two-dimensional PCA figure. ### build_pose_grid_pca_figure { #aria_nbv.vin.diagnostics.plotting.build_pose_grid_pca_figure } ```python vin.diagnostics.plotting.build_pose_grid_pca_figure( pos_grid, *, pos_proj, max_points=8000, color_by='radius', show_axes=True, axis_scale=0.5, ) ``` Project positional grid embeddings to 2D via PCA. ### build_pose_grid_slices_figure { #aria_nbv.vin.diagnostics.plotting.build_pose_grid_slices_figure } ```python vin.diagnostics.plotting.build_pose_grid_slices_figure(pos_grid, *, axis, index) ``` Plot 2D slices of the normalized position grid (x/y/z channels). ### build_pose_vec_histogram { #aria_nbv.vin.diagnostics.plotting.build_pose_vec_histogram } ```python vin.diagnostics.plotting.build_pose_vec_histogram( pose_vec, *, dim_index, num_bins=60, log1p_counts=False, ) ``` Histogram of a single pose vector component. ### build_scene_field_evidence_figures { #aria_nbv.vin.diagnostics.plotting.build_scene_field_evidence_figures } ```python vin.diagnostics.plotting.build_scene_field_evidence_figures( debug, *, channel_names, occ_threshold=0.5, max_points=40000, ) ``` Plot scene-field channels (field_in) as world-space scatter plots. ### build_se3_closure_figure { #aria_nbv.vin.diagnostics.plotting.build_se3_closure_figure } ```python vin.diagnostics.plotting.build_se3_closure_figure( candidate_poses_world_cam, reference_pose_world_rig, *, bins=50, log1p_counts=False, ) ``` Check closure consistency of T_world_cam = T_world_rig_ref * T_rig_ref_cam. ### build_semidense_cnn_grid_figure { #aria_nbv.vin.diagnostics.plotting.build_semidense_cnn_grid_figure } ```python vin.diagnostics.plotting.build_semidense_cnn_grid_figure( points_world, *, p3d_cameras, candidate_index, grid_size, max_points, ) ``` Visualize the projection grid maps used by the VIN v3 semidense CNN. ### build_semidense_cnn_grid_maps { #aria_nbv.vin.diagnostics.plotting.build_semidense_cnn_grid_maps } ```python vin.diagnostics.plotting.build_semidense_cnn_grid_maps( points_world, *, p3d_cameras, candidate_index, grid_size, max_points, ) ``` Compute semidense grid maps that match VIN v3's tiny-CNN inputs. ### build_semidense_projection_feature_figure { #aria_nbv.vin.diagnostics.plotting.build_semidense_projection_feature_figure } ```python vin.diagnostics.plotting.build_semidense_projection_feature_figure( points_world, *, p3d_cameras, candidate_index, grid_size, max_points, semidense_obs_count_max, semidense_inv_dist_std_min, semidense_inv_dist_std_p95, ) ``` Visualize semidense projection feature maps for a candidate. ### build_semidense_projection_feature_maps { #aria_nbv.vin.diagnostics.plotting.build_semidense_projection_feature_maps } ```python vin.diagnostics.plotting.build_semidense_projection_feature_maps( points_world, *, p3d_cameras, candidate_index, grid_size, max_points, semidense_obs_count_max, semidense_inv_dist_std_min, semidense_inv_dist_std_p95, ) ``` Compute semidense projection feature maps for a candidate view. ### build_semidense_projection_figure { #aria_nbv.vin.diagnostics.plotting.build_semidense_projection_figure } ```python vin.diagnostics.plotting.build_semidense_projection_figure( points_world, *, p3d_cameras, candidate_index, max_points=20000, show_frustum=False, frustum_scale=None, frustum_color='#ff4d4d', ) ``` Plot semidense points colored by whether they project inside the candidate view. ### build_valid_fraction_figure { #aria_nbv.vin.diagnostics.plotting.build_valid_fraction_figure } ```python vin.diagnostics.plotting.build_valid_fraction_figure(debug) ``` Plot candidate centers colored by validity fraction. ### build_voxel_frame_figure { #aria_nbv.vin.diagnostics.plotting.build_voxel_frame_figure } ```python vin.diagnostics.plotting.build_voxel_frame_figure(debug, *, axis_scale=0.5) ``` Plot voxel grid bounds and voxel axes in world coordinates. ### build_voxel_inbounds_figure { #aria_nbv.vin.diagnostics.plotting.build_voxel_inbounds_figure } ```python vin.diagnostics.plotting.build_voxel_inbounds_figure( candidate_poses_world_cam, t_world_voxel, voxel_extent, *, bins=50, ) ``` Plot candidate in-bounds ratios and normalized coordinate distributions. ### build_voxel_roundtrip_figure { #aria_nbv.vin.diagnostics.plotting.build_voxel_roundtrip_figure } ```python vin.diagnostics.plotting.build_voxel_roundtrip_figure( debug, *, num_points=2000, log1p_counts=False, ) ``` Visualize metric residuals from a voxel-to-world-to-voxel round trip.