Detailed Description

Class for generating 3D bounding box key points based on anchors.

This class implements the AnchorHandler interface for generating key points from 3D box anchors and providing anchor projection functionality.

Definition at line 34 of file sources/gst-plugins/gst-nvdsvideotemplate/ds-sparse4d/postprocess/sparse_box_3d_generator.h.

Inheritance diagram for SparseBox3DKeyPointsGenerator:

[legend]

Collaboration diagram for SparseBox3DKeyPointsGenerator:

[legend]

Public Member Functions
	SparseBox3DKeyPointsGenerator (int embed_dims=256, int num_learnable_pts=0, const std::vector< std::tuple< float, float, float >> &fix_scale={{0.0, 0.0, 0.0}})
	Constructor for SparseBox3DKeyPointsGenerator. More...

void	init_weight ()
	Initialize weights with proper initialization schemes. More...

std::variant< torch::Tensor, std::tuple< torch::Tensor, std::vector< torch::Tensor > > >	forward (const torch::Tensor &anchor, const torch::Tensor &instance_feature=torch::Tensor(), const std::vector< torch::Tensor > &T_cur2temp_list={}, const torch::Tensor &cur_timestamp=torch::Tensor(), const std::vector< torch::Tensor > &temp_timestamps={})
	Forward pass to generate key points from anchors. More...

std::vector< torch::Tensor >	anchor_projection (const torch::Tensor &anchors, const std::vector< torch::Tensor > &transforms, const std::vector< torch::Tensor > &time_intervals) override
	Project anchors using transformation matrices and time intervals. More...

	SparseBox3DKeyPointsGenerator (int embed_dims=256, int num_learnable_pts=0, const std::vector< std::tuple< float, float, float >> &fix_scale={{0.0, 0.0, 0.0}})
	Constructor for SparseBox3DKeyPointsGenerator. More...

void	init_weight ()
	Initialize weights with proper initialization schemes. More...

std::variant< torch::Tensor, std::tuple< torch::Tensor, std::vector< torch::Tensor > > >	forward (const torch::Tensor &anchor, const torch::Tensor &instance_feature=torch::Tensor(), const std::vector< torch::Tensor > &T_cur2temp_list={}, const torch::Tensor &cur_timestamp=torch::Tensor(), const std::vector< torch::Tensor > &temp_timestamps={})
	Forward pass to generate key points from anchors. More...

std::vector< torch::Tensor >	anchor_projection (const torch::Tensor &anchors, const std::vector< torch::Tensor > &transforms, const std::vector< torch::Tensor > &time_intervals) override
	Project anchors using transformation matrices and time intervals. More...

Static Public Member Functions
static std::vector< torch::Tensor >	anchor_projection_static (const torch::Tensor &anchor, const std::vector< torch::Tensor > &T_src2dst_list, const torch::Tensor &src_timestamp=torch::Tensor(), const std::vector< torch::Tensor > &dst_timestamps={}, const std::vector< torch::Tensor > &time_intervals={})
	Static method to project anchors using various parameters. More...

static torch::Tensor	distance (const torch::Tensor &anchor)
	Calculate Euclidean distance of anchors from origin in 2D. More...

static std::vector< torch::Tensor >	anchor_projection_static (const torch::Tensor &anchor, const std::vector< torch::Tensor > &T_src2dst_list, const torch::Tensor &src_timestamp=torch::Tensor(), const std::vector< torch::Tensor > &dst_timestamps={}, const std::vector< torch::Tensor > &time_intervals={})
	Static method to project anchors using various parameters. More...

static torch::Tensor	distance (const torch::Tensor &anchor)
	Calculate Euclidean distance of anchors from origin in 2D. More...

Constructor & Destructor Documentation

◆ SparseBox3DKeyPointsGenerator() [1/2]

SparseBox3DKeyPointsGenerator::SparseBox3DKeyPointsGenerator	(	int	embed_dims = `256`,
		int	num_learnable_pts = `0`,
		const std::vector< std::tuple< float, float, float >> &	fix_scale = `{{0.0, 0.0, 0.0}}`
	)

Constructor for SparseBox3DKeyPointsGenerator.

Parameters

embed_dims	Dimension of embedding features
num_learnable_pts	Number of learnable key points
fix_scale	Fixed scale values for key points (array of 3D points)

◆ SparseBox3DKeyPointsGenerator() [2/2]

SparseBox3DKeyPointsGenerator::SparseBox3DKeyPointsGenerator	(	int	embed_dims = `256`,
		int	num_learnable_pts = `0`,
		const std::vector< std::tuple< float, float, float >> &	fix_scale = `{{0.0, 0.0, 0.0}}`
	)

Constructor for SparseBox3DKeyPointsGenerator.

Parameters

embed_dims	Dimension of embedding features
num_learnable_pts	Number of learnable key points
fix_scale	Fixed scale values for key points (array of 3D points)

Member Function Documentation

◆ anchor_projection() [1/2]

std::vector<torch::Tensor> SparseBox3DKeyPointsGenerator::anchor_projection	(	const torch::Tensor &	anchors,
		const std::vector< torch::Tensor > &	transforms,
		const std::vector< torch::Tensor > &	time_intervals
	)

overridevirtual

Project anchors using transformation matrices and time intervals.

Parameters

anchors	Input anchors to project
transforms	Transformation matrices for coordinate transforms
time_intervals	Time intervals for temporal projection

Returns: Vector of projected anchor tensors

Implements AnchorHandler.

◆ anchor_projection() [2/2]

std::vector<torch::Tensor> SparseBox3DKeyPointsGenerator::anchor_projection	(	const torch::Tensor &	anchors,
		const std::vector< torch::Tensor > &	transforms,
		const std::vector< torch::Tensor > &	time_intervals
	)

overridevirtual

Project anchors using transformation matrices and time intervals.

Parameters

anchors	Input anchors to project
transforms	Transformation matrices for coordinate transforms
time_intervals	Time intervals for temporal projection

Returns: Vector of projected anchor tensors

Implements AnchorHandler.

◆ anchor_projection_static() [1/2]

static std::vector<torch::Tensor> SparseBox3DKeyPointsGenerator::anchor_projection_static	(	const torch::Tensor &	anchor,
		const std::vector< torch::Tensor > &	T_src2dst_list,
		const torch::Tensor &	src_timestamp = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	dst_timestamps = `{}`,
		const std::vector< torch::Tensor > &	time_intervals = `{}`
	)

static

Static method to project anchors using various parameters.

Parameters

anchor	Input anchor tensor
T_src2dst_list	List of transformation matrices
src_timestamp	Optional source timestamp
dst_timestamps	Optional destination timestamps
time_intervals	Optional time intervals

Returns: Vector of projected anchor tensors

◆ anchor_projection_static() [2/2]

static std::vector<torch::Tensor> SparseBox3DKeyPointsGenerator::anchor_projection_static	(	const torch::Tensor &	anchor,
		const std::vector< torch::Tensor > &	T_src2dst_list,
		const torch::Tensor &	src_timestamp = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	dst_timestamps = `{}`,
		const std::vector< torch::Tensor > &	time_intervals = `{}`
	)

static

Static method to project anchors using various parameters.

Parameters

anchor	Input anchor tensor
T_src2dst_list	List of transformation matrices
src_timestamp	Optional source timestamp
dst_timestamps	Optional destination timestamps
time_intervals	Optional time intervals

Returns: Vector of projected anchor tensors

◆ distance() [1/2]

static torch::Tensor SparseBox3DKeyPointsGenerator::distance ( const torch::Tensor & anchor )

static

Calculate Euclidean distance of anchors from origin in 2D.

Parameters

anchor Input anchor tensor

Returns: Tensor of distances

◆ distance() [2/2]

static torch::Tensor SparseBox3DKeyPointsGenerator::distance ( const torch::Tensor & anchor )

static

Calculate Euclidean distance of anchors from origin in 2D.

Parameters

anchor Input anchor tensor

Returns: Tensor of distances

◆ forward() [1/2]

std::variant<torch::Tensor, std::tuple<torch::Tensor, std::vector<torch::Tensor> > > SparseBox3DKeyPointsGenerator::forward	(	const torch::Tensor &	anchor,
		const torch::Tensor &	instance_feature = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	T_cur2temp_list = `{}`,
		const torch::Tensor &	cur_timestamp = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	temp_timestamps = `{}`
	)

Forward pass to generate key points from anchors.

Parameters

anchor	Anchor tensors [batch_size, num_anchor, dim]
instance_feature	Optional instance features for learnable points
T_cur2temp_list	Optional transformation matrices
cur_timestamp	Optional current timestamp
temp_timestamps	Optional temporal timestamps

Returns: key points tensor or tuple of key points and temporal key points

◆ forward() [2/2]

std::variant<torch::Tensor, std::tuple<torch::Tensor, std::vector<torch::Tensor> > > SparseBox3DKeyPointsGenerator::forward	(	const torch::Tensor &	anchor,
		const torch::Tensor &	instance_feature = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	T_cur2temp_list = `{}`,
		const torch::Tensor &	cur_timestamp = `torch::Tensor()`,
		const std::vector< torch::Tensor > &	temp_timestamps = `{}`
	)

Forward pass to generate key points from anchors.

Parameters

anchor	Anchor tensors [batch_size, num_anchor, dim]
instance_feature	Optional instance features for learnable points
T_cur2temp_list	Optional transformation matrices
cur_timestamp	Optional current timestamp
temp_timestamps	Optional temporal timestamps

Returns: key points tensor or tuple of key points and temporal key points

◆ init_weight() [1/2]

void SparseBox3DKeyPointsGenerator::init_weight ( )

Initialize weights with proper initialization schemes.

◆ init_weight() [2/2]

void SparseBox3DKeyPointsGenerator::init_weight ( )

Initialize weights with proper initialization schemes.

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

sources/gst-plugins/gst-nvdsvideotemplate/ds-sparse4d/postprocess/sparse_box_3d_generator.h

NVIDIA DeepStream SDK API Reference

9.2 Release

Detailed Description

Public Member Functions

Static Public Member Functions

Constructor & Destructor Documentation

◆ SparseBox3DKeyPointsGenerator() [1/2]

◆ SparseBox3DKeyPointsGenerator() [2/2]

Member Function Documentation

◆ anchor_projection() [1/2]

◆ anchor_projection() [2/2]

◆ anchor_projection_static() [1/2]

◆ anchor_projection_static() [2/2]

◆ distance() [1/2]

◆ distance() [2/2]

◆ forward() [1/2]

◆ forward() [2/2]

◆ init_weight() [1/2]

◆ init_weight() [2/2]