AsyncPipeline Class Reference

#include <async_pipeline.h>

Collaboration diagram for AsyncPipeline:

Public Member Functions
	AsyncPipeline (std::unique_ptr< ModelBase > &&modelInstance, const ModelConfig &config, ov::Core &core)
void	waitForData (bool shouldKeepOrder=true)
bool	isReadyToProcess ()
void	waitForTotalCompletion ()
virtual int64_t	submitData (const InputData &inputData, const std::shared_ptr< MetaData > &metaData)
virtual std::unique_ptr< ResultBase >	getResult (bool shouldKeepOrder=true)
PerformanceMetrics	getInferenceMetircs ()
PerformanceMetrics	getPreprocessMetrics ()
PerformanceMetrics	getPostprocessMetrics ()

Protected Member Functions
virtual InferenceResult	getInferenceResult (bool shouldKeepOrder)

Protected Attributes
std::unique_ptr< RequestsPool >	requestsPool
std::unordered_map< int64_t, InferenceResult >	completedInferenceResults
ov::CompiledModel	compiledModel
std::mutex	mtx
std::condition_variable	condVar
int64_t	inputFrameId = 0
int64_t	outputFrameId = 0
std::exception_ptr	callbackException = nullptr
std::unique_ptr< ModelBase >	model
PerformanceMetrics	inferenceMetrics
PerformanceMetrics	preprocessMetrics
PerformanceMetrics	postprocessMetrics

Detailed Description

This is base class for asynchronous pipeline Derived classes should add functions for data submission and output processing

Constructor & Destructor Documentation

AsyncPipeline()

AsyncPipeline::AsyncPipeline	(	std::unique_ptr< ModelBase > &&	modelInstance,
		const ModelConfig &	config,
		ov::Core &	core
	)

Loads model and performs required initialization

Parameters

modelInstance	pointer to model object. Object it points to should not be destroyed manually after passing pointer to this function.
config	- fine tuning configuration for model
core	- reference to ov::Core instance to use. If it is omitted, new instance of ov::Core will be created inside.

Member Function Documentation

getInferenceResult()

InferenceResult AsyncPipeline::getInferenceResult ( bool  shouldKeepOrder )

protectedvirtual

Returns processed result, if available

Parameters

shouldKeepOrder

if true, function will return processed data sequentially, keeping original frames order (as they were submitted). Otherwise, function will return processed data in random order.

Returns

InferenceResult with processed information or empty InferenceResult (with negative frameID) if there's no any results yet.

getResult()

std::unique_ptr< ResultBase > AsyncPipeline::getResult ( bool  shouldKeepOrder = true )

virtual

Gets available data from the queue

Parameters

shouldKeepOrder

if true, function will treat results as ready only if next sequential result (frame) is ready (so results can be extracted in the same order as they were submitted). Otherwise, function will return if any result is ready.

isReadyToProcess()

bool AsyncPipeline::isReadyToProcess ( )

inline

Returns

true if there's available infer requests in the pool and next frame can be submitted for processing, false otherwise.

submitData()

int64_t AsyncPipeline::submitData ( const InputData &  inputData, const std::shared_ptr< MetaData > &  metaData  )

virtual

Submits data to the model for inference

Parameters

inputData	- input data to be submitted
metaData	- shared pointer to metadata container. Might be null. This pointer will be passed through pipeline and put to the final result structure.

Returns

-1 if image cannot be scheduled for processing (there's no free InferRequest available). Otherwise returns unique sequential frame ID for this particular request. Same frame ID will be written in the result structure.

waitForData()

void AsyncPipeline::waitForData

(

bool

shouldKeepOrder = true

)

Waits until either output data becomes available or pipeline allows to submit more input data.

Parameters

shouldKeepOrder

if true, function will treat results as ready only if next sequential result (frame) is ready (so results can be extracted in the same order as they were submitted). Otherwise, function will return if any result is ready.

waitForTotalCompletion()

void AsyncPipeline::waitForTotalCompletion ( )

inline

Waits for all currently submitted requests to be completed.

---

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

• src/common/pipelines/include/pipelines/async_pipeline.h
• src/common/pipelines/src/async_pipeline.cpp