vfe

Dynamic voxel feature encoder shared by the voxel detectors (Voxel Mamba, LION).

A packed-format port of the DynamicVoxelVFE from gwenzhang/Voxel-Mamba.

Classes:

PFNLayer –

Pillar feature net layer (the reference's PFNLayerV2): linear + norm + ReLU with a per-voxel max-pool.
DynamicMeanVFE –

Dynamic mean voxel feature encoder for the Voxel Mamba detector.

PFNLayer ¶

PFNLayer(in_channels: int, out_channels: int, last: bool)

Bases: Module

Pillar feature net layer (the reference's PFNLayerV2): linear + norm + ReLU with a per-voxel max-pool.

Non-final layers halve their output width and concatenate the pooled feature back onto every point; the final layer returns the pooled per-voxel feature directly.

Parameters:

in_channels (int) –

Input feature channels.
out_channels (int) –

Target output channels (halved internally for non-final layers).
last (bool) –

Whether this is the final PFN layer.

Shape

Input: \((N, C_\text{in})\) point features and \((N,)\) voxel index.
Output: \((N, C')\) for non-final layers, \((M, C_\text{out})\) for the final layer.

DynamicMeanVFE ¶

DynamicMeanVFE(
    in_channels: int,
    num_filters: Sequence[int],
    voxel_size: Sequence[float],
    point_cloud_range: Sequence[float],
    grid_size: Sequence[int],
)

Bases: Module

Dynamic mean voxel feature encoder for the Voxel Mamba detector.

Points are assigned to voxels on the fly (no fixed points-per-voxel), augmented with the per-voxel cluster-mean offset and voxel-center offset, then encoded by a stack of linear + norm + ReLU PFN layers whose per-voxel max-pool produces one feature vector per voxel.

Reference implementation: gwenzhang/Voxel-Mamba (DynamicVoxelVFE).

Parameters:

in_channels (int) –

Raw point feature channels including xyz (e.g. \(5\) for Waymo \(x, y, z, \text{intensity}, \text{elongation}\)).
num_filters (Sequence[int]) –

Output width of each PFN layer; the last entry is the voxel feature dim.
voxel_size (Sequence[float]) –

Voxel size \((v_x, v_y, v_z)\).
point_cloud_range (Sequence[float]) –

Range \((x_\min, y_\min, z_\min, x_\max, y_\max, z_\max)\).
grid_size (Sequence[int]) –

Voxel grid extent \((n_x, n_y, n_z)\).

Shape

Input: \((N, C_\text{in})\) point features and \((N,)\) batch index.
Output: \((M, C_\text{out})\) voxel features and \((M, 4)\) voxel coords.