Enums

class fvdb.SmoothingMode(*values)

Laplacian smoothing mode used to de-staircase a signed distance field in fvdb.Grid.reinitialize_sdf() / fvdb.Grid.retopologize_sdf() (and their fvdb.GridBatch counterparts).

The number of smoothing passes is controlled separately by the smooth argument; this enum selects which umbrella-Laplacian flow each pass applies. Values mirror the C++ fvdb::detail::ops::SmoothingMode enum.

MEAN_CURVATURE = 0

each pass moves every voxel toward the average of its 6 face neighbours. Effective at removing staircase artifacts but shrinks the surface (volume loss) if over-applied.

Type:

Mean-curvature flow

TAUBIN = 1

alternates a positive (shrinking) and a slightly larger negative (inflating) Laplacian step per pass, de-staircasing with much less volume loss.

Type:

Volume-preserving Taubin smoothing

class fvdb.ConvolutionPhasePolicy(*values)

Kernel phase convention used by a convolution plan.

TORCH_SAME_PHASE = 'torch_same_phase'

Use PyTorch padding=floor((kernel_size - 1) / 2) phase.

class fvdb.ConvolutionTopologyPolicy(*values)

Policy controlling the finite output topology of a convolution plan.

COMPLETE = 'complete'

Generate the complete, uncropped structural topology.

RESTRICTED = 'restricted'

Evaluate the convolution relation only on an explicit target grid.

class fvdb.ConvolutionTopologyProvenance(*values)

How a convolution plan’s finite topology was obtained.

EXACT_TRANSPOSE = 'exact_transpose'

Reversed directly from another plan’s stored finite edge set.

EXPLICIT_TARGET = 'explicit_target'

Restricted to a target grid supplied by the caller.

GENERATED = 'generated'

Generated from the source grid using the canonical convolution relation.