SimaVoxelizeOptions Struct

Construction options for a SimaVoxelize Node. More...

Declaration

struct simaai::neat::SimaVoxelizeOptions { ... }

Included Headers

#include <SimaVoxelize.h>

Public Member Attributes Index

std::array< float, 3 >	voxel_size {}
	Metres per cell in x, y and z. z must span the whole range: one vertical cell. More...
std::array< float, 6 >	point_cloud_range {}
	x_lo, y_lo, z_lo, x_hi, y_hi, z_hi in metres. Points outside are dropped. More...
SimaKernelQuantization	quantization
	Quantization of the decorated FP32 features on store (INT8 output only). More...
TensorDType	output_dtype = TensorDType::Int8
	Element type of decorated: TensorDType::Int8 or TensorDType::BFloat16. More...
int	max_pillars = 0
	Pillar capacity M. Pillars beyond it are dropped. More...
int	max_points_per_pillar = 0
	Points-per-pillar capacity T. Points beyond it are dropped. More...
SimaKernelTarget	target = SimaKernelTarget::A65
	Execution target; no fallback. More...

Description

Construction options for a SimaVoxelize Node.

The input Sample holds two tensors, in order: FP32 points [N,4] (x, y, z and the sweep time offset) and INT32 point_count [1]. Only the leading point_count points are read. The output Sample holds decorated [1,M,T,10] (INT8 or BF16, per output_dtype), INT32 coordinates [M,2] (row, column) and INT32 count [1].

A BF16 decorated tensor is not quantized: each binary32 channel is rounded to nearest-even BF16, NaN stores the canonical 32704, and quantization must stay the identity (multiplier 1, zero point 0, nearest-even).

Definition at line 39 of file SimaVoxelize.h.

Public Member Attributes

max_pillars

int simaai::neat::SimaVoxelizeOptions::max_pillars = 0

Pillar capacity M. Pillars beyond it are dropped.

Definition at line 49 of file SimaVoxelize.h.

max_points_per_pillar

int simaai::neat::SimaVoxelizeOptions::max_points_per_pillar = 0

Points-per-pillar capacity T. Points beyond it are dropped.

Definition at line 51 of file SimaVoxelize.h.

output_dtype

TensorDType simaai::neat::SimaVoxelizeOptions::output_dtype = TensorDType::Int8

Element type of decorated: TensorDType::Int8 or TensorDType::BFloat16.

Definition at line 47 of file SimaVoxelize.h.

point_cloud_range

std::array<float, 6> simaai::neat::SimaVoxelizeOptions::point_cloud_range {}

x_lo, y_lo, z_lo, x_hi, y_hi, z_hi in metres. Points outside are dropped.

Definition at line 43 of file SimaVoxelize.h.

quantization

SimaKernelQuantization simaai::neat::SimaVoxelizeOptions::quantization

Quantization of the decorated FP32 features on store (INT8 output only).

Definition at line 45 of file SimaVoxelize.h.

target

SimaKernelTarget simaai::neat::SimaVoxelizeOptions::target = SimaKernelTarget::A65

Execution target; no fallback.

Definition at line 53 of file SimaVoxelize.h.

voxel_size

std::array<float, 3> simaai::neat::SimaVoxelizeOptions::voxel_size {}

Metres per cell in x, y and z. z must span the whole range: one vertical cell.

Definition at line 41 of file SimaVoxelize.h.

---

The documentation for this struct was generated from the following file:

• SimaVoxelize.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.15.0.