UNeuralNetwork

Choose your operating system:

Windows

macOS

Linux

Inheritance Hierarchy

UObjectBase

UObjectBaseUtility

UObject

UNeuralNetwork

References

Module	NeuralNetworkInference
Header	/Engine/Plugins/Experimental/NeuralNetworkInference/Source/NeuralNetworkInference/Public/NeuralNetwork.h
Include	#include "NeuralNetwork.h"

Syntax

UCLASS(BlueprintType)
class UNeuralNetwork : public UObject

Variables

	Name	Description
ENeuralDeviceTy...	DeviceType	The neural device type of the network.
ENeuralDeviceTy...	InputDeviceType	It defines whether Run() will expect the input data in CPU memory (Run will upload the memory to the GPU) or GPU memory (no upload needed).
FString	ModelFullFilePath	Original model file path from which this UNeuralNetwork was loaded from.
ENeuralDeviceTy...	OutputDeviceType	It defines whether Run() will return the output data in CPU memory (Run will download the memory to the CPU) or GPU (no download needed).
ENeuralSynchron...	SynchronousMode	It defines whether UNeuralNetwork::Run() will block the thread until completed (Synchronous), or whether it will run on a background thread, not blocking the calling thread (Asynchronous).
ENeuralThreadMo...	ThreadModeDelegateForAsyncRunCompleted	If SynchronousMode is Asynchronous, this variable will define whether the callback delegate is called from the game thread (highly recommended) or from any available thread (not fully thread safe).

Constructors

	Name	Description
	UNeuralNetwork()	Default constructor that initializes the internal class variables.

Destructors

	Name	Description
	~UNeuralNetwork()	Destructor defined in case this class is ever inherited.

Functions

	Name	Description
	DECLARE_DELEGATE ( FOnAsyncRunCompleted )	FOnAsyncRunCompleted is the delegate that gets triggered when an asynchronous inference has been completed (i.e., Run() was called with NeuralSynchronousMode == ENeuralSynchronousMode::Asynchronous).
TObjectPtr< ...	GetAndMaybeCreateAssetImportData()
TObjectPtr< ...	GetAssetImportData()	Internal and Editor-only functions not needed by the user.
ENeuralBackE...	GetBackEnd()	Internal functions that should not be used by the user.
ENeuralDevic...	GetDeviceType()	Getter and setter functions for DeviceType, InputDeviceType, and OutputDeviceType:
void *	GetInputDataPointerMutable ( const int32 InTensorIndex )
ENeuralDevic...	GetInputDeviceType()
FNeuralStati...	GetInputMemoryTransferStats()
const FNeura...	GetInputTensor ( const int32 InTensorIndex )	GetInputTensor() and GetOutputTensor() return a const (read-only) reference of the input or output FNeuralTensor(s) of the network, respectively.
int64	GetInputTensorNumber()	GetInputTensorNumber() and GetOutputTensorNumber() return the number of input or output tensors of this network, respectively.
float	GetLastRunTimeMSec()	Statistics-related functions:
FOnAsyncRunC...	GetOnAsyncRunCompletedDelegate()
ENeuralDevic...	GetOutputDeviceType()
const FNeura...	GetOutputTensor ( const int32 InTensorIndex )
int64	GetOutputTensorNumber()
FNeuralStati...	GetRunStatistics()
ENeuralSynch...	GetSynchronousMode()	Getter and setter functions for SynchronousMode:
ENeuralThrea...	GetThreadModeDelegateForAsyncRunCompleted()	Getter and setter functions for ThreadModeDelegateForAsyncRunCompleted:
void	InputTensorsToGPU ( const TArray< int32 >& InTensorInd... )	Non-computationally-efficient functions meant to be used only for debugging purposes, but should never be used on highly performant systems:
bool	IsGPUSupported()	Whether GPU execution (i.e., SetDeviceType(ENeuralDeviceType::GPU)) is supported for this platform:
bool	IsLoaded()	It returns whether a network is currently loaded. It is equivalent to the output of Load().
bool	Load ( const FString& InModelFilePath )	Load() + SetInputFromArrayCopy() + Run() is the simplest way to load an ONNX file, set the input tensor(s), and run inference on it.
bool	Load ( TArray< uint8 >& InModelReadFromFi... )	Load() + SetInputFromArrayCopy() + Run() is the simplest way to load an ONNX file, set the input tensor(s), and run inference on it.
bool	Load ( TArray< FNeuralTensor >& InTensors, const TArray< FNeuralTensor* >..., const TArray< FNeuralTensor* >..., const TArray< TSharedPtr< class FNe... )	Internal function not needed by the user.
void	OutputTensorsToCPU ( const TArray< int32 >& InTensorInd... )
void	ResetStats()
void	Run()	Load() + SetInputFromArrayCopy() + Run() is the simplest way to load an ONNX file, set the input tensor(s), and run inference on it.
void	SetDeviceType ( const ENeuralDeviceType InDeviceTyp..., const ENeuralDeviceType InInputDevi..., const ENeuralDeviceType InOutputDev... )
void	SetInputFromArrayCopy ( const TArray< float >& InArray, const int32 InTensorIndex )	SetInputFromArrayCopy(), SetInputFromVoidPointerCopy(), and GetInputDataPointerMutable() are the only functions that allow modifying the network input tensor(s) values:
void	SetInputFromVoidPointerCopy ( const void*const InVoidPtr, const int32 InTensorIndex )
void	SetSynchronousMode ( const ENeuralSynchronousMode InSync... )
void	SetThreadModeDelegateForAsyncRunCompleted ( const ENeuralThreadMode InThreadMod... )

Classes

	Name	Description
	FImplBackEndUEOnly	FImplBackEndUEOnly

Enums

	Name	Description
	ENeuralBackEnd	Internal enum class that should not be used by the user.

Deprecated Functions

		Name	Description
	ENeuralBackE...	GetBackEndForCurrentPlatform()	Do not use, ENeuralBackEnd will be removed in future versions and only UEAndORT back end will be supported.
	bool	SetBackEnd ( const ENeuralBackEnd InBackEnd )	Do not use, ENeuralBackEnd will be removed in future versions and only UEAndORT back end will be supported.