Detailed Description

InstanceBank class for managing and tracking object instances across frames.

This class implements functionality to maintain and update feature representations of object instances across time, supporting object tracking and temporal association.

Definition at line 61 of file sources/gst-plugins/gst-nvdsvideotemplate/ds-sparse4d/postprocess/instance_bank.h.

Inheritance diagram for InstanceBank:

[legend]

Collaboration diagram for InstanceBank:

[legend]

Public Member Functions
	InstanceBank (int num_anchor, int embed_dims, const torch::Tensor &anchor, AnchorHandler *anchor_handler=nullptr, int num_temp_instances=0, float default_time_interval=0.5, float confidence_decay=0.6, bool anchor_grad=true, bool feat_grad=true, float max_time_interval=2.0, int reid_dims=-1, int gpu_id=-1)
	Constructor for InstanceBank. More...

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

void	reset ()
	Reset internal state. More...

std::tuple< torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor >	get (int batch_size, const MetaData &metas, const DNMetaData *dn_metas=nullptr)
	Get features, anchors and time intervals for the current batch. More...

std::tuple< torch::Tensor, torch::Tensor >	update (torch::Tensor instance_feature, torch::Tensor anchor, torch::Tensor confidence)
	Update instance features and anchors. More...

void	cache (torch::Tensor instance_feature, torch::Tensor anchor, torch::Tensor confidence, const MetaData metas=nullptr, const std::vector< torch::Tensor > feature_maps=nullptr)
	Cache features, anchors and metadata for temporal association. More...

torch::Tensor	get_instance_id (torch::Tensor confidence, torch::Tensor anchor=torch::Tensor(), float threshold=-1.0)
	Get instance IDs based on confidence scores. More...

void	update_instance_id (torch::Tensor instance_id_tensor, torch::Tensor confidence=torch::Tensor())
	Update instance IDs based on confidence scores. More...

torch::Tensor	getMask () const
	Get the mask tensor. More...

	InstanceBank (int num_anchor, int embed_dims, const torch::Tensor &anchor, AnchorHandler *anchor_handler=nullptr, int num_temp_instances=0, float default_time_interval=0.5, float confidence_decay=0.6, bool anchor_grad=true, bool feat_grad=true, float max_time_interval=2.0, int reid_dims=-1, int gpu_id=-1)
	Constructor for InstanceBank. More...

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

void	reset ()
	Reset internal state. More...

std::tuple< torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor >	get (int batch_size, const MetaData &metas, const DNMetaData *dn_metas=nullptr)
	Get features, anchors and time intervals for the current batch. More...

std::tuple< torch::Tensor, torch::Tensor >	update (torch::Tensor instance_feature, torch::Tensor anchor, torch::Tensor confidence)
	Update instance features and anchors. More...

void	cache (torch::Tensor instance_feature, torch::Tensor anchor, torch::Tensor confidence, const MetaData metas=nullptr, const std::vector< torch::Tensor > feature_maps=nullptr)
	Cache features, anchors and metadata for temporal association. More...

torch::Tensor	get_instance_id (torch::Tensor confidence, torch::Tensor anchor=torch::Tensor(), float threshold=-1.0)
	Get instance IDs based on confidence scores. More...

void	update_instance_id (torch::Tensor instance_id_tensor, torch::Tensor confidence=torch::Tensor())
	Update instance IDs based on confidence scores. More...

torch::Tensor	getMask () const
	Get the mask tensor. More...

Constructor & Destructor Documentation

◆ InstanceBank() [1/2]

InstanceBank::InstanceBank	(	int	num_anchor,
		int	embed_dims,
		const torch::Tensor &	anchor,
		AnchorHandler *	anchor_handler = `nullptr`,
		int	num_temp_instances = `0`,
		float	default_time_interval = `0.5`,
		float	confidence_decay = `0.6`,
		bool	anchor_grad = `true`,
		bool	feat_grad = `true`,
		float	max_time_interval = `2.0`,
		int	reid_dims = `-1`,
		int	gpu_id = `-1`
	)

Constructor for InstanceBank.

Parameters

num_anchor	Maximum number of anchors to use
embed_dims	Dimension of the embedding features
anchor	Initial anchor tensor
anchor_handler	Optional handler for anchor projection
num_temp_instances	Number of temporary instances to track
default_time_interval	Default time interval between frames
confidence_decay	Decay factor for confidence scores
anchor_grad	Whether to enable gradients for anchor parameters
feat_grad	Whether to enable gradients for feature parameters
max_time_interval	Maximum time interval to consider valid
reid_dims	Optional re-identification feature dimension
gpu_id	Device ID to use for computations

◆ InstanceBank() [2/2]

InstanceBank::InstanceBank	(	int	num_anchor,
		int	embed_dims,
		const torch::Tensor &	anchor,
		AnchorHandler *	anchor_handler = `nullptr`,
		int	num_temp_instances = `0`,
		float	default_time_interval = `0.5`,
		float	confidence_decay = `0.6`,
		bool	anchor_grad = `true`,
		bool	feat_grad = `true`,
		float	max_time_interval = `2.0`,
		int	reid_dims = `-1`,
		int	gpu_id = `-1`
	)

Constructor for InstanceBank.

Parameters

num_anchor	Maximum number of anchors to use
embed_dims	Dimension of the embedding features
anchor	Initial anchor tensor
anchor_handler	Optional handler for anchor projection
num_temp_instances	Number of temporary instances to track
default_time_interval	Default time interval between frames
confidence_decay	Decay factor for confidence scores
anchor_grad	Whether to enable gradients for anchor parameters
feat_grad	Whether to enable gradients for feature parameters
max_time_interval	Maximum time interval to consider valid
reid_dims	Optional re-identification feature dimension
gpu_id	Device ID to use for computations

Member Function Documentation

◆ cache() [1/2]

void InstanceBank::cache	(	torch::Tensor	instance_feature,
		torch::Tensor	anchor,
		torch::Tensor	confidence,
		const MetaData *	metas = `nullptr`,
		const std::vector< torch::Tensor > *	feature_maps = `nullptr`
	)

Cache features, anchors and metadata for temporal association.

Parameters

instance_feature	Instance features to cache
anchor	Anchors to cache
confidence	Confidence scores
metas	Optional metadata
feature_maps	Optional feature maps

◆ cache() [2/2]

void InstanceBank::cache	(	torch::Tensor	instance_feature,
		torch::Tensor	anchor,
		torch::Tensor	confidence,
		const MetaData *	metas = `nullptr`,
		const std::vector< torch::Tensor > *	feature_maps = `nullptr`
	)

Cache features, anchors and metadata for temporal association.

Parameters

instance_feature	Instance features to cache
anchor	Anchors to cache
confidence	Confidence scores
metas	Optional metadata
feature_maps	Optional feature maps

◆ get() [1/2]

std::tuple<torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor> InstanceBank::get	(	int	batch_size,
		const MetaData &	metas,
		const DNMetaData *	dn_metas = `nullptr`
	)

Get features, anchors and time intervals for the current batch.

Parameters

batch_size	Size of the batch
metas	Metadata for the current frames
dn_metas	Optional denoising metadata

Returns: Tuple of tensors: instance_feature, anchor, cached_feature, cached_anchor, time_interval, reid_feature

◆ get() [2/2]

std::tuple<torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor, torch::Tensor> InstanceBank::get	(	int	batch_size,
		const MetaData &	metas,
		const DNMetaData *	dn_metas = `nullptr`
	)

Get features, anchors and time intervals for the current batch.

Parameters

batch_size	Size of the batch
metas	Metadata for the current frames
dn_metas	Optional denoising metadata

Returns: Tuple of tensors: instance_feature, anchor, cached_feature, cached_anchor, time_interval, reid_feature

◆ get_instance_id() [1/2]

torch::Tensor InstanceBank::get_instance_id	(	torch::Tensor	confidence,
		torch::Tensor	anchor = `torch::Tensor()`,
		float	threshold = `-1.0`
	)

Get instance IDs based on confidence scores.

Parameters

confidence	Confidence scores for instances
anchor	Optional anchor tensor
threshold	Optional confidence threshold (negative value means no threshold)

Returns: Tensor of instance IDs

◆ get_instance_id() [2/2]

torch::Tensor InstanceBank::get_instance_id	(	torch::Tensor	confidence,
		torch::Tensor	anchor = `torch::Tensor()`,
		float	threshold = `-1.0`
	)

Get instance IDs based on confidence scores.

Parameters

confidence	Confidence scores for instances
anchor	Optional anchor tensor
threshold	Optional confidence threshold (negative value means no threshold)

Returns: Tensor of instance IDs

◆ getMask() [1/2]

torch::Tensor InstanceBank::getMask ( ) const

inline

Get the mask tensor.

Returns: The internal mask tensor

Definition at line 169 of file sources/gst-plugins/gst-nvdsvideotemplate/ds-sparse4d/postprocess/instance_bank.h.

◆ getMask() [2/2]

torch::Tensor InstanceBank::getMask ( ) const

inline

Get the mask tensor.

Returns: The internal mask tensor

Definition at line 169 of file 9.2/sources/gst-plugins/gst-nvdsvideotemplate/ds-sparse4d/postprocess/instance_bank.h.

◆ init_weight() [1/2]

void InstanceBank::init_weight ( )

Initialize weights with proper initialization schemes.

◆ init_weight() [2/2]

void InstanceBank::init_weight ( )

Initialize weights with proper initialization schemes.

◆ reset() [1/2]

void InstanceBank::reset ( )

Reset internal state.

◆ reset() [2/2]

void InstanceBank::reset ( )

Reset internal state.

◆ update() [1/2]

std::tuple<torch::Tensor, torch::Tensor> InstanceBank::update	(	torch::Tensor	instance_feature,
		torch::Tensor	anchor,
		torch::Tensor	confidence
	)

Update instance features and anchors.

Parameters

instance_feature	Current instance features
anchor	Current anchors
confidence	Confidence scores

Returns: Updated instance features and anchors

◆ update() [2/2]

std::tuple<torch::Tensor, torch::Tensor> InstanceBank::update	(	torch::Tensor	instance_feature,
		torch::Tensor	anchor,
		torch::Tensor	confidence
	)

Update instance features and anchors.

Parameters

instance_feature	Current instance features
anchor	Current anchors
confidence	Confidence scores

Returns: Updated instance features and anchors

◆ update_instance_id() [1/2]

void InstanceBank::update_instance_id	(	torch::Tensor	instance_id_tensor,
		torch::Tensor	confidence = `torch::Tensor()`
	)

Update instance IDs based on confidence scores.

Parameters

instance_id_tensor	Current instance IDs
confidence	Optional confidence scores

◆ update_instance_id() [2/2]

void InstanceBank::update_instance_id	(	torch::Tensor	instance_id_tensor,
		torch::Tensor	confidence = `torch::Tensor()`
	)

Update instance IDs based on confidence scores.

Parameters

instance_id_tensor	Current instance IDs
confidence	Optional confidence scores

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

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

NVIDIA DeepStream SDK API Reference

9.2 Release

Detailed Description

Public Member Functions

Constructor & Destructor Documentation

◆ InstanceBank() [1/2]

◆ InstanceBank() [2/2]

Member Function Documentation

◆ cache() [1/2]

◆ cache() [2/2]

◆ get() [1/2]

◆ get() [2/2]

◆ get_instance_id() [1/2]

◆ get_instance_id() [2/2]

◆ getMask() [1/2]

◆ getMask() [2/2]

◆ init_weight() [1/2]

◆ init_weight() [2/2]

◆ reset() [1/2]

◆ reset() [2/2]

◆ update() [1/2]

◆ update() [2/2]

◆ update_instance_id() [1/2]

◆ update_instance_id() [2/2]