NN::OrtNeuralNet Class Reference

A class that implements the NeuralNet interface using ONNX Runtime. More...

#include <OrtNeuralNet.h>

Inheritance diagram for NN::OrtNeuralNet:

Public Member Functions
	OrtNeuralNet (const std::string &modelPath, bool useCuda=false, int gpuId=0)
	Construct a new OrtNeuralNet object.
tl::expected< std::vector< Tensor >, NeuralError >	infer (const Tensor &input) override
	Perform inference on the given input Tensor using the ONNX model.
size_t	inputCount () const override
	Get the number of inputs of the neural network.
size_t	outputCount () const override
	Get the number of outputs of the neural network.
Shape	inputShape (size_t idx=0) const override
	Get the input shape of the neural network.
Shape	outputShape (size_t idx=0) const override
	Get the output shape of the neural network.
std::string	inputName (size_t idx=0) const override
	Get the name of the input tensor.
std::string	outputName (size_t idx=0) const override
	Get the name of the output tensor.
int	gpuId () const override
	Get the GPU ID used by the neural network if it is configured to use GPU.
Public Member Functions inherited from NN::NeuralNet
tl::expected< std::vector< Tensor >, NeuralError >	operator() (const Tensor &input)
	Call the infer method with the given input tensor.

Detailed Description

A class that implements the NeuralNet interface using ONNX Runtime.

This class is responsible for loading an ONNX model, performing inference, and converting between Tensor and ONNX Runtime's Ort::Value. It supports both CPU and GPU execution, depending on the model and the environment setup.

Note

DO NOT use this class directly in your code. Instead, use the NeuralNetFactory and NeuralNet interface to interact with neural networks.

Usage:

// Create an OrtNeuralNet instance
auto neuralNet = NeuralNetFactory::create("model.onnx");
if (!neuralNet) {
    std::cerr << "Failed to create neural network: " << neuralNet.error() << std::endl;
    return -1;
}
 
// Create a Tensor with the correct shape and data type
Shape inputShape = neuralNet->inputShape(); // assume it returns {1, 3, 224, 224} for a single image input
cv::Mat inputMat(224, 224, CV_32FC3, cv::Scalar(0.0f)); // Example input
Tensor inputTensor = Tensor::fromCvMat(inputMat).value(); // Convert cv::Mat to Tensor
 
// Perform inference
auto output = neuralNet->infer(inputTensor);
if (!output) {
    std::cerr << "Inference failed: " << output.error() << std::endl;
    return -1;
}
Tensor outputTensor = output.value();

Constructor & Destructor Documentation

OrtNeuralNet()

NN::OrtNeuralNet::OrtNeuralNet	(	const std::string &	modelPath,
		bool	useCuda = false,
		int	gpuId = 0
	)

Construct a new OrtNeuralNet object.

Parameters

modelPath	The path to the ONNX model file.
useCuda	Whether to use CUDA for GPU execution. Default is false (CPU).
gpuId	The ID of the GPU to use if CUDA is enabled. Default is 0.

Note

DO NOT use this constructor directly in your code. Instead, use the NeuralNetFactory to create an instance of this class.

Member Function Documentation

gpuId()

int NN::OrtNeuralNet::gpuId ( ) const

overridevirtual

Get the GPU ID used by the neural network if it is configured to use GPU.

Returns

int The GPU ID, or -1 if the neural network does not use GPU.

Implements NN::NeuralNet.

infer()

tl::expected< std::vector< NN::Tensor >, NN::NeuralError > NN::OrtNeuralNet::infer ( const Tensor &  input )

overridevirtual

Perform inference on the given input Tensor using the ONNX model.

Parameters

input

The input Tensor to the neural network. This tensor must have the correct shape and data type expected by the model.

Returns

tl::expected<std::vector<Tensor>, NeuralError> The output Tensor or an error object.

Implements NN::NeuralNet.

inputCount()

size_t NN::OrtNeuralNet::inputCount ( ) const

overridevirtual

Get the number of inputs of the neural network.

Returns

size_t The number of inputs.

Implements NN::NeuralNet.

inputName()

std::string NN::OrtNeuralNet::inputName ( size_t  idx = 0 ) const

overridevirtual

Get the name of the input tensor.

Parameters

idx	The index of the input to query (default is 0). Valid indices are in the range [0, inputCount()).

Returns

std::string The name of the input tensor.

Implements NN::NeuralNet.

inputShape()

NN::Shape NN::OrtNeuralNet::inputShape ( size_t  idx = 0 ) const

overridevirtual

Get the input shape of the neural network.

Parameters

idx	The index of the input to query (default is 0). Valid indices are in the range [0, inputCount()).

Returns

Shape The shape of the input tensor expected by the model.

Implements NN::NeuralNet.

outputCount()

size_t NN::OrtNeuralNet::outputCount ( ) const

overridevirtual

Get the number of outputs of the neural network.

Returns

size_t The number of outputs.

Implements NN::NeuralNet.

outputName()

std::string NN::OrtNeuralNet::outputName ( size_t  idx = 0 ) const

overridevirtual

Get the name of the output tensor.

Parameters

idx	The index of the output to query (default is 0). Valid indices are in the range [0, outputCount()).

Returns

std::string The name of the output tensor.

Implements NN::NeuralNet.

outputShape()

NN::Shape NN::OrtNeuralNet::outputShape ( size_t  idx = 0 ) const

overridevirtual

Get the output shape of the neural network.

Parameters

idx	The index of the output to query (default is 0). Valid indices are in the range [0, outputCount()).

Returns

Shape The shape of the output tensor produced by the model.

Implements NN::NeuralNet.

---

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

• iVS3D/src/iVS3D-neuralNet/include/OrtNeuralNet.h
• iVS3D/src/iVS3D-neuralNet/src/OrtNeuralNet.cpp