Typedefs
template<typename ImagePointType>
using	ImagesPatchValuesAndJacobiansEvaluator
template<typename ImagePointType>
using	ImagesPatchValuesEvaluator

Functions
template<class ImagePointType>
std::vector< torch::Tensor >	GenerateOutputs (const std::vector< itk::ImpactModelConfiguration > &modelConfig, const std::vector< ImagePointType > &fixedPoints, const std::vector< std::vector< std::vector< std::vector< float > > > > &patchIndex, const std::vector< torch::Tensor > subsetsOfFeatures, torch::Device device, const ImpactTensorUtils::ImagesPatchValuesEvaluator< ImagePointType > &imagesPatchValuesEvaluator)
template<typename ImagePointType>
std::vector< torch::Tensor >	GenerateOutputsAndJacobian (const std::vector< itk::ImpactModelConfiguration > &modelConfig, const std::vector< ImagePointType > &fixedPoints, const std::vector< std::vector< std::vector< std::vector< float > > > > &patchIndex, std::vector< torch::Tensor > subsetsOfFeatures, std::vector< torch::Tensor > fixedOutputsTensor, torch::Device device, std::vector< std::unique_ptr< ImpactLoss::Loss > > &losses, const ImpactTensorUtils::ImagesPatchValuesAndJacobiansEvaluator< ImagePointType > &imagesPatchValuesAndJacobiansEvaluator)
template<typename TImage, typename FeaturesMaps, typename InterpolatorType, typename FeaturesImageType>
std::vector< FeaturesMaps >	GetFeaturesMaps (typename TImage::ConstPointer image, typename InterpolatorType::Pointer interpolator, const std::vector< itk::ImpactModelConfiguration > &modelsConfiguration, torch::Device device, std::vector< unsigned int > pca, std::vector< torch::Tensor > &principalComponents, const std::function< void(typename TImage::ConstPointer, torch::Tensor &, const std::string &)> &writeInputImage, const std::function< typename TImage::PointType(const typename TImage::PointType &)> &transformPoint=nullptr)
std::vector< torch::Tensor >	GetModelOutputsExample (std::vector< itk::ImpactModelConfiguration > &modelsConfig, const std::string &modelType, torch::Device device)
std::vector< std::vector< float > >	GetPatchIndex (const itk::ImpactModelConfiguration &modelConfiguration, std::mt19937 &randomGenerator, unsigned int dimension)
template<typename TImage, typename TInterpolator>
torch::Tensor	ImageToTensor (typename TImage::ConstPointer image, typename TInterpolator::Pointer interpolator, const std::vector< float > &voxelSize, const std::function< typename TImage::PointType(const typename TImage::PointType &)> &transformPoint)
template<typename TImage, typename TFeatureImage>
TFeatureImage::Pointer	TensorToImage (typename TImage::ConstPointer image, torch::Tensor layers)

Typedef Documentation

◆ ImagesPatchValuesAndJacobiansEvaluator

template<typename ImagePointType>

using ImpactTensorUtils::ImagesPatchValuesAndJacobiansEvaluator

Initial value:

 std::function<torch::Tensor(const ImagePointType &,
                                                                           torch::Tensor &,
                                                                           const std::vector<std::vector<float>> &,
                                                                           const std::vector<int64_t> &,
                                                                           int)>

Definition at line 174 of file ImpactTensorUtils.h.

◆ ImagesPatchValuesEvaluator

template<typename ImagePointType>

using ImpactTensorUtils::ImagesPatchValuesEvaluator

Initial value:

std::function<

torch::Tensor(const ImagePointType &, const std::vector<std::vector<float>> &, const std::vector<int64_t> &)>

Definition at line 146 of file ImpactTensorUtils.h.

Function Documentation

◆ GenerateOutputs()

template<class ImagePointType>

std::vector< torch::Tensor > ImpactTensorUtils::GenerateOutputs	(	const std::vector< itk::ImpactModelConfiguration > &	modelConfig,
		const std::vector< ImagePointType > &	fixedPoints,
		const std::vector< std::vector< std::vector< std::vector< float > > > > &	patchIndex,
		const std::vector< torch::Tensor >	subsetsOfFeatures,
		torch::Device	device,
		const ImpactTensorUtils::ImagesPatchValuesEvaluator< ImagePointType > &	imagesPatchValuesEvaluator )

Computes feature outputs for all patches using each model.

This function computes feature outputs for each patch by running a forward pass through the model and applying optional feature sub-selection.

Parameters

modelConfig	Vector of model configurations (dimensions, patch sizes, voxel sizes, etc.).
fixedPoints	Central points on the fixed image where features are computed.
patchIndex	Indices defining the local patches around each point.
subsetsOfFeatures	Tensors containing indices of feature channels to select.
device	The device (CPU or GPU) to perform the computation on.
imagesPatchValuesEvaluator	Callable that evaluate image values.

Returns: A vector of tensors containing the computed feature outputs for each patch.

◆ GenerateOutputsAndJacobian()

template<typename ImagePointType>

std::vector< torch::Tensor > ImpactTensorUtils::GenerateOutputsAndJacobian	(	const std::vector< itk::ImpactModelConfiguration > &	modelConfig,
		const std::vector< ImagePointType > &	fixedPoints,
		const std::vector< std::vector< std::vector< std::vector< float > > > > &	patchIndex,
		std::vector< torch::Tensor >	subsetsOfFeatures,
		std::vector< torch::Tensor >	fixedOutputsTensor,
		torch::Device	device,
		std::vector< std::unique_ptr< ImpactLoss::Loss > > &	losses,
		const ImpactTensorUtils::ImagesPatchValuesAndJacobiansEvaluator< ImagePointType > &	imagesPatchValuesAndJacobiansEvaluator )

Computes feature outputs and their spatial Jacobians for image registration.

This function calculates the feature outputs and their corresponding Jacobians at given points to aid in optimization. The Jacobians are used for gradient computation during image registration.

Parameters

modelConfig	Vector of model configurations (dimensions, patch sizes, voxel sizes, etc.).
fixedPoints	Central points on the fixed image where features are computed.
patchIndex	Indices defining the local patches around each point for feature extraction.
subsetsOfFeatures	Tensors containing indices of feature channels to select.
fixedOutputsTensor	Tensor containing feature outputs from the fixed image.
device	The device (CPU or GPU) to perform the computation on.
losses	Vector of loss objects to be updated incrementally during optimization.
imagesPatchValuesAndJacobiansEvaluator	Callable that evaluate image values and Jacobians.

Returns: A vector of tensors with computed feature outputs.

◆ GetFeaturesMaps()

template<typename TImage, typename FeaturesMaps, typename InterpolatorType, typename FeaturesImageType>

std::vector< FeaturesMaps > ImpactTensorUtils::GetFeaturesMaps	(	typename TImage::ConstPointer	image,
		typename InterpolatorType::Pointer	interpolator,
		const std::vector< itk::ImpactModelConfiguration > &	modelsConfiguration,
		torch::Device	device,
		std::vector< unsigned int >	pca,
		std::vector< torch::Tensor > &	principalComponents,
		const std::function< void(typename TImage::ConstPointer, torch::Tensor &, const std::string &)> &	writeInputImage,
		const std::function< typename TImage::PointType(const typename TImage::PointType &)> &	transformPoint = nullptr )

Applies one or more models to an image to extract feature maps.

This function extracts feature maps from an image using one or more models. Optionally, it can export the resampled input image for debugging purposes.

Parameters

image	The input image from which features are extracted.
interpolator	The interpolator used for resampling.
modelsConfiguration	The configuration of the models used for feature extraction.
device	The device (CPU or GPU) to perform the computation on.
pca	The number of principal components for dimensionality reduction.
principalComponents	A vector to store the computed principal components.
writeInputImage	Optional function to export the resampled input image for debugging.
transformPoint	ptional function to transform a point, used for geometric transformations on the moving image.

Returns: A vector of ITK feature images, one per layer and model.

◆ GetModelOutputsExample()

std::vector< torch::Tensor > ImpactTensorUtils::GetModelOutputsExample	(	std::vector< itk::ImpactModelConfiguration > &	modelsConfig,
		const std::string &	modelType,
		torch::Device	device )

Tests the configuration of each model by generating outputs from dummy input.

This function validates the compatibility of TorchScript models and checks the output structure (e.g., number of layers, spatial shape, channels). It ensures that models are properly loaded and executable during initialization.

Parameters

modelsConfig	Vector of model configurations.
modelType	The type of the model being tested (fixed, moving) for logging errors.
device	The device (CPU or GPU) to perform the computation on.

Returns: A vector of tensors with the generated outputs from dummy input.

◆ GetPatchIndex()

std::vector< std::vector< float > > ImpactTensorUtils::GetPatchIndex	(	const itk::ImpactModelConfiguration &	modelConfiguration,
		std::mt19937 &	randomGenerator,
		unsigned int	dimension )

Computes patch index offsets around a center point based on model configuration.

This function generates the offsets for local patches (e.g., 5x5x5) around a center point using the model's configuration. It helps in extracting features from local neighborhoods.

Parameters

modelConfiguration	Configuration of the model, including patch size and voxel size.
randomGenerator	Random generator used for randomizing 2D patch slices in a 3D volume.
dimension	The image dimension (2D or 3D).

Returns: A vector of offsets for the patch around the center point.

◆ ImageToTensor()

template<typename TImage, typename TInterpolator>

torch::Tensor ImpactTensorUtils::ImageToTensor	(	typename TImage::ConstPointer	image,
		typename TInterpolator::Pointer	interpolator,
		const std::vector< float > &	voxelSize,
		const std::function< typename TImage::PointType(const typename TImage::PointType &)> &	transformPoint )

Converts an ITK image to a Torch tensor using physical spacing.

Parameters

voxelSize	Target voxel size in mm, used for resampling the image.
transformPoint	Optional point-wise transform to apply to the moving image.

Returns: A Torch tensor representing the resampled ITK image.

◆ TensorToImage()

template<typename TImage, typename TFeatureImage>

TFeatureImage::Pointer ImpactTensorUtils::TensorToImage	(	typename TImage::ConstPointer	image,
		torch::Tensor	layers )

Converts a tensor (C×D×H×W) to a multi-channel ITK image. Converts the given tensor to an ITK image, preserving the original image metadata (origin, spacing, direction).

Parameters

image	The original input image used to retrieve metadata.
layers	The tensor representing the feature layers.

Returns: A multi-channel ITK image constructed from the tensor.

Typedefs

Functions

Typedef Documentation

◆ ImagesPatchValuesAndJacobiansEvaluator

◆ ImagesPatchValuesEvaluator

Function Documentation

◆ GenerateOutputs()

◆ GenerateOutputsAndJacobian()

◆ GetFeaturesMaps()

◆ GetModelOutputsExample()

◆ GetPatchIndex()

◆ ImageToTensor()

◆ TensorToImage()