types

Shared type aliases, TypedDict definitions, and enums used across the package.

Classes:

Boxes3D –

Packed 3D boxes with class and scene indices (e.g. detection ground truth, PyG batch layout).
Detection3D –

Packed 3D detections: Boxes3D plus a per-box confidence score (a model's decode output).
FeaturesDict –

Features of one encoder stage on a packed point cloud, as returned in the return_intermediates list.
PooledFeaturesDict –

FeaturesDict of a stage pooled by clustering, keeping the map into the next coarser stage.
MessagePassingParams –

Keyword arguments a message-passing layer forwards to its MessagePassing base class.
StrEnum –

Enum whose members are plain strings, so they compare, print and hash like their value.

Bases: TypedDict

Packed 3D boxes with class and scene indices (e.g. detection ground truth, PyG batch layout).

Attributes:

boxes (Tensor) –

Boxes \((N, 7)\) of \((c_x, c_y, c_z, d_x, d_y, d_z, \theta)\).
labels (Tensor) –

Per-box class, shape \((N,)\).
batch (Tensor) –

Per-box scene index, shape \((N,)\).
ignore_mask (NotRequired[Tensor]) –

Per-box ignore mask, shape \((N,)\).

Bases: Boxes3D

Packed 3D detections: Boxes3D plus a per-box confidence score (a model's decode output).

Attributes:

boxes (Tensor) –

Boxes \((N, 7)\) of \((c_x, c_y, c_z, d_x, d_y, d_z, \theta)\).
labels (Tensor) –

Per-box class, shape \((N,)\).
batch (Tensor) –

Per-box scene index, shape \((N,)\).
ignore_mask (NotRequired[Tensor]) –

Per-box ignore mask, shape \((N,)\).
scores (Tensor) –

Per-box confidence score, shape \((N,)\).
class_probs (NotRequired[Tensor]) –

Per-box class probabilities, shape \((N, C)\), emitted by detectors whose eval protocol expands every kept box across all classes (e.g. VoteNet / 3DETR).
velocity (NotRequired[Tensor]) –

Per-box BEV velocity \((v_x, v_y)\), shape \((N, 2)\), emitted by detectors whose head regresses velocity (the nuScenes heads).

Bases: TypedDict

Features of one encoder stage on a packed point cloud, as returned in the return_intermediates list.

Attributes:

x (Tensor) –

Stage features, shape \((N_s, C_s)\).
batch (Tensor) –

Per-point batch index, shape \((N_s,)\).
pos (NotRequired[Tensor]) –

Positions of the stage's points, shape \((N_s, 3)\).
pos_grid (NotRequired[Tensor]) –

Integer grid coordinates of the stage's voxels, shape \((N_s, 3)\).

FeaturesDict of a stage pooled by clustering, keeping the map into the next coarser stage.

Attributes:

pooling_inverse (NotRequired[Tensor]) –

Index of each point's cluster in the next stage, shape \((N_s,)\).

Bases: TypedDict

Keyword arguments a message-passing layer forwards to its MessagePassing base class.

Bases: str, Enum

Enum whose members are plain strings, so they compare, print and hash like their value.