gstnvinferserver Namespace Reference

Data Structures
class	GstNvInferServerImpl
	Class of the nvinferserver element implementation. More...
struct	GstNvInferServerObjectHistory
	Holds the inference information/history for one object based on its tracking ID. More...
struct	GstNvInferServerProperties
	Holds the configuration information from the nvinferserver element properties. More...
struct	GstNvInferServerSourceInfo
	Holds information about the detected objects in the specific input source. More...

Typedefs
using	SharedObjHistory = std::shared_ptr< GstNvInferServerObjectHistory >
using	WeakObjHistory = std::weak_ptr< GstNvInferServerObjectHistory >
typedef std::unordered_map< uint64_t, SharedObjHistory >	GstNvInferServerObjectHistoryMap
	Map for maintaining inference history for objects based on their tracking ids: Object ID : Object History. More...
using	RawOutputCallback = std::function< void(void *, NvDsInferNetworkInfo &, NvDsInferLayerInfo *, uint32_t, uint32_t)>
	Function for raw output callback . More...

Functions
void	attachDetectionMetadata (NvDsFrameMeta *frameMeta, NvDsObjectMeta *parentObj, const NvDsInferDetectionOutput &detection_output, float scaleX, float scaleY, uint32_t offsetLeft, uint32_t offsetTop, uint32_t roiLeft, uint32_t roiTop, uint32_t roiWidth, uint32_t roiHeight, uint32_t imageWidth, uint32_t imageHeight, uint32_t uniqueId, const ic::PluginControl &config)
	Attach detection metadata for the objects in a frame. More...
void	attachClassificationMetadata (NvDsObjectMeta *objMeta, NvDsFrameMeta *frameMeta, NvDsRoiMeta *roiMeta, const InferClassificationOutput &objInfo, uint32_t uniqueId, const std::string &classifierType, uint32_t imageWidth, uint32_t imageHeight)
	Attach the classification output as NvDsClassifierMeta. More...
void	mergeClassificationOutput (InferClassificationOutput &cache, const InferClassificationOutput &newRes)
	Merge the object history with the new classification result. More...
void	attachSegmentationMetadata (NvDsObjectMeta *objMeta, NvDsFrameMeta *frameMeta, NvDsRoiMeta *roiMeta, const NvDsInferSegmentationOutput &segmentation_output, dsis::SharedIBatchBuffer &buf)
	Attach the segmentation output as user metadata. More...
void	attachTensorOutputMeta (NvDsObjectMeta *objMeta, NvDsFrameMeta *frameMeta, NvDsRoiMeta *roiMeta, uint32_t uniqueId, const std::vector< dsis::SharedIBatchBuffer > &tensors, uint32_t batchIdx, const NvDsInferNetworkInfo &inputInfo, bool maintainAspectRatio)
	Attaches the raw tensor output to the GstBuffer as metadata. More...
void	attachFullTensorOutputMeta (NvDsBatchMeta *batchMeta, uint32_t uniqueId, const std::vector< dsis::SharedIBatchBuffer > &tensors, const NvDsInferNetworkInfo &inputInfo)
	Attach the full inference output tensors to the batch metadata. More...

Typedef Documentation

GstNvInferServerObjectHistoryMap

typedef std::unordered_map<uint64_t, SharedObjHistory> gstnvinferserver::GstNvInferServerObjectHistoryMap

Map for maintaining inference history for objects based on their tracking ids: Object ID : Object History.

Definition at line 97 of file gstnvinferserver_impl.h.

RawOutputCallback

using gstnvinferserver::RawOutputCallback = typedef std::function<void(void *, NvDsInferNetworkInfo&, NvDsInferLayerInfo*, uint32_t, uint32_t)>

Function for raw output callback .

Function for raw output callback (void *gstBuf, NvDsInferNetworkInfo *network_info, NvDsInferLayerInfo *layers_info, uint32_t num_layers, uint32_t batch_size)

Definition at line 135 of file gstnvinferserver_impl.h.

SharedObjHistory

using gstnvinferserver::SharedObjHistory = typedef std::shared_ptr<GstNvInferServerObjectHistory>

Definition at line 90 of file gstnvinferserver_impl.h.

WeakObjHistory

using gstnvinferserver::WeakObjHistory = typedef std::weak_ptr<GstNvInferServerObjectHistory>

Definition at line 91 of file gstnvinferserver_impl.h.

Function Documentation

attachClassificationMetadata()

void gstnvinferserver::attachClassificationMetadata	(	NvDsObjectMeta *	objMeta,
		NvDsFrameMeta *	frameMeta,
		NvDsRoiMeta *	roiMeta,
		const InferClassificationOutput &	objInfo,
		uint32_t	uniqueId,
		const std::string &	classifierType,
		uint32_t	imageWidth,
		uint32_t	imageHeight
	)

Attach the classification output as NvDsClassifierMeta.

This functions creates a new metadata of the type NvDsClassifierMeta and attaches it to the object metadata or ROI metadata. If processing on full frames, a new object metadata is created and attached. The display text for the object metadata is updated base on the label in the classification result.

Parameters

[in,out]	objMeta	Pointer to the object metadata of the inference input.
[in,out]	frameMeta	Pointer to the frame metadata of the inference input.
[in,out]	roiMeta	Pointer to the ROI metadata for the inference input.
[in]	objInfo	The classification output from the inference context.
[in]	uniqueId	Unique ID of the GIE instance.
[in]	classifierType	The classifier type of the nvinferserver element.
[in]	imageWidth	The frame width used as bounding box size when inferencing on full frames.
[in]	imageHeight	The frame height used as bounding box size when inferencing on full frames.

attachDetectionMetadata()

void gstnvinferserver::attachDetectionMetadata	(	NvDsFrameMeta *	frameMeta,
		NvDsObjectMeta *	parentObj,
		const NvDsInferDetectionOutput &	detection_output,
		float	scaleX,
		float	scaleY,
		uint32_t	offsetLeft,
		uint32_t	offsetTop,
		uint32_t	roiLeft,
		uint32_t	roiTop,
		uint32_t	roiWidth,
		uint32_t	roiHeight,
		uint32_t	imageWidth,
		uint32_t	imageHeight,
		uint32_t	uniqueId,
		const ic::PluginControl &	config
	)

Attach detection metadata for the objects in a frame.

This function attaches metadata for the objects detected in a frame. It adds a new object metadata, NvDsObjectMeta, to the metadata of the input frame. The bounding box parameters of the detected objects are scaled and shifted as per the actual frame dimensions using the provided scale and offset values.

Parameters

[in,out]	frameMeta	Pointer to the NvDSFrameMeta structure for the frame.
[in]	parentObj	Pointer to the NvDsObjMeta structure for the parent object under inference. It is NULL in case of primary detector.
[in]	detection_output	The detection output from the inference context.
[in]	scaleX	Horizontal scaling ratio that was applied to the input to get network input dimensions.
[in]	scaleY	vertical scaling ratio that was applied to the input to get network input dimensions.
[in]	offsetLeft	Left offset corresponding to the left padding applied to the scaled input.
[in]	offsetTop	Top offset corresponding to the top padding added to the scaled input.
[in]	roiLeft	Left pixel coordinate of the input ROI in the parent frame.
[in]	roiTop	Top pixel coordinate of the input ROI in the parent frame.
[in]	roiWidth	Width of the ROI in the parent frame.
[in]	roiHeight	Height of the ROI in the parent frame.
[in]	imageWidth	Width of the frames in the input batch buffer.
[in]	imageHeight	Height of the frames in the input batch buffer.
[in]	uniqueId	Unique ID of the GIE instance.
[in]	config	The configuration settings used for the GStreamer element.

attachFullTensorOutputMeta()

void gstnvinferserver::attachFullTensorOutputMeta	(	NvDsBatchMeta *	batchMeta,
		uint32_t	uniqueId,
		const std::vector< dsis::SharedIBatchBuffer > &	tensors,
		const NvDsInferNetworkInfo &	inputInfo
	)

Attach the full inference output tensors to the batch metadata.

This functions attaches the output tensors for the entire batch as a user metadata of type NVDSINFER_TENSOR_OUTPUT_META.

Parameters

[in,out]	batchMeta	Pointer to the bach metadata of the inference input.
[in]	uniqueId	Unique ID of the GIE instance.
[in]	tensors	The batch buffer array of the output tensors.
[in]	inputInfo	Dimensions of the input layer for the network.

attachSegmentationMetadata()

void gstnvinferserver::attachSegmentationMetadata	(	NvDsObjectMeta *	objMeta,
		NvDsFrameMeta *	frameMeta,
		NvDsRoiMeta *	roiMeta,
		const NvDsInferSegmentationOutput &	segmentation_output,
		dsis::SharedIBatchBuffer &	buf
	)

Attach the segmentation output as user metadata.

This function adds the segmentation output as user metadata of the input buffer. The function adds the user metadata to the ROI metadata if present. Otherwise, if object metadata is present, the user data is added to object metadata. Otherwise it is added to the frame metadata.

Parameters

[in,out]	objMeta	Pointer to the object metadata of the inference input.
[in,out]	frameMeta	Pointer to the frame metadata of the inference input.
[in,out]	roiMeta	Pointer to the ROI metadata for the inference input.
[in]	segmentation_output	The segmentation output from the inference context.
[in]	buf	Pointer to the batch buffer containing the segmentation output. This is saved as the priv_data in the user metadata.

attachTensorOutputMeta()

void gstnvinferserver::attachTensorOutputMeta	(	NvDsObjectMeta *	objMeta,
		NvDsFrameMeta *	frameMeta,
		NvDsRoiMeta *	roiMeta,
		uint32_t	uniqueId,
		const std::vector< dsis::SharedIBatchBuffer > &	tensors,
		uint32_t	batchIdx,
		const NvDsInferNetworkInfo &	inputInfo,
		bool	maintainAspectRatio
	)

Attaches the raw tensor output to the GstBuffer as metadata.

This function attaches the output tensors for each frame in the input batch to the correspoinding ROI metadata or or object metadata or frame metadata.

Parameters

[in,out]	objMeta	Pointer to the object metadata of the inference input.
[in,out]	frameMeta	Pointer to the frame metadata of the inference input.
[in,out]	roiMeta	Pointer to the ROI metadata for the inference input.
[in]	uniqueId	Unique ID of the GIE instance.
[in]	tensors	The batch buffer array of the output tensors.
[in]	batchIdx	Index of the frame within the inference batch.
[in]	inputInfo	Dimensions of the input layer for the network.
[in]	maintainAspectRatio	maintain_aspect_ratio configuration setting value.

mergeClassificationOutput()

void gstnvinferserver::mergeClassificationOutput	(	InferClassificationOutput &	cache,
		const InferClassificationOutput &	newRes
	)

Merge the object history with the new classification result.

Given an object history, merge the new classification results with the previous cached results. This can be used to improve the results of classification when re-inferencing over time. Currently, the function just uses the latest results.

Parameters

[in,out]	cache	The cached classification output in the object history.
[in]	newRes	New classification output for the object.