ImFusion::ML Namespace Reference

Namespace containing all the classes relevant for the data processing and the execution of machine learning models. More...

Detailed Description

Namespace containing all the classes relevant for the data processing and the execution of machine learning models.

See also

ImFusionML

Classes
class	AddPositionAsChannelAlgorithm
	Algorithm to add a new channel to a given image containing each pixel's position. More...
struct	ParamException
	Special type for handling exceptions during parameter configuration. More...
class	AdvancedParameter
	The AdvancedParameter class extends the Parameter class by providing some additional functionalities used in ML::Operation and ML::ImageROISampler classes, like specifying whether a parameter is required for the object to work, and automatically logging an error message in case the user doesn't provide it. More...
struct	Box
	Bounding Box for ML tasks Since bounding boxes are axis aligned by definition, a Box is represented by its center and its extent. More...
class	BoundingBoxSet
	Class for managing sets of bounding boxes within a BoundingBoxElement The class is meant to be used in parallel with SharedImageSet. More...
struct	CudaComputeCapability
	Helper struct to store the CUDA compute capability. More...
class	CudaDeviceManager
	Singleton class that helps getting a compatible CUDA device. More...
struct	DataElementException
	Custom exception thrown by some DataElement. More...
class	DataElement
	Main interface for data elements used for wrapping other ImFusion types and use them in a machine learning specific workflow. More...
struct	DataItemException
	Custom exception thrown by DataItem. More...
class	DataItem
	Class for holding a map of heterogeneous DataElements, see DataElement. More...
class	DataItemDisplayHandler
	DataDisplayHandler for DataItems. More...
struct	DataLoaderException
	Custom exception thrown by DataLoader. More...
class	DataLoader
	Interface for all data loaders, establishing the mechanism for loading and processing data. More...
class	DataReader
	Base class for data readers. More...
struct	DataLoaderSpecs
class	Dataset
	Class for creating an iterable dataset by chaining data loading and transforming operations executed in a lazy fashion. More...
class	Engine
	Generic interface for machine learning models serialized by specific frameworks (PyTorch, ONNX, etc.). More...
class	EngineConfiguration
class	GenerateBoxSegmentationTrainingDataAlgorithm
	Algorithm to generate learning data for box segmentation network. More...
class	Singleton
	Base class for singletons. More...
class	Registry
class	Factory
	Generic factory class for registering polymorphic types Usage: More...
class	GroupToDataItemAlgorithm
	Groups a list of Data into a DataItem. More...
struct	RegionOfInterest
	Helper struct to represent a region of interest. More...
struct	ImageSamplersException
	Custom exception. More...
class	ImageROISampler
	Interface for samplers used during training of machine learning models. More...
class	KeypointExtractionAlgorithm
	Algorithm for extracting keypoints from blobs representing the probability distribution of the keypoint location. More...
class	KeypointSet
	Class for managing sets of keypoints within a KeypointsElement The class is meant to be used in parallel with SharedImageSet. More...
class	KeypointSetIoAlgorithm
	Io Algorithm to load/save KeypointSet. More...
class	BoundingBoxSetIoAlgorithm
	Io Algorithm to load/save BoundingBoxSet. More...
class	LandmarkPredictionAlgorithm
	Algorithm to predict landmarks in an image. More...
class	ClassificationLearningEvaluation
	Class for evaluation of a binary classification method. More...
class	MultiClassificationLearningEvaluation
	Class for evaluation of a multi-label classification method. More...
class	RegressionLearningEvaluation
	Class for evaluation of a regression method. More...
class	LocalConvolutionalNetworkAlgorithm
	Algorithm to perform a Local (possibly augmented) prediction using a fully convolutional network around a central point. More...
class	LocalizeSegmentAlgorithm
	Localize & Segment Algorithm Runs an initial low resolution pixelwise segmentation to find a bounding box. More...
struct	MLModelException
	Custom exception thrown by MachineLearningModel. More...
class	MachineLearningModel
	Class for managing and executing a machine learning model on generic input data. More...
class	MachineLearningModelAlgorithm
	Generic algorithm to apply a machine learning model. More...
class	MarkovChain
class	MatrixBasedOperation
	Extension of the Operation class for operations based on a geometric transformation using a matrix (e.g. More...
class	Metric
	The Metric interface is a key part of our pipeline for standardized evaluations of a machine learning pipeline, or an algorithm in general. More...
struct	MetricException
	Custom exception for metrics. More...
class	MetricAlgorithm
	Algorithm to compute metrics from a generic list of data The selected metric need to be set via the configure method: DataList algoInput = {labelMap1.get(), labelMap2.get()}; MetricAlgorithm metricAlgo(algoInput); Properties algoProps; algoProps.setParam("metricName", "DiceMetric"); metricAlgo.configure(&algoProps); metricAlgo.compute(); std::vector<Metric::Record> results = metricAlgo.metricOutput();. More...
class	ImagewiseClassificationMetrics
	Compute the confusion matrix for a multi-class classification. More...
class	DiceMetric
	Compute the Dice score between a label map and a ground truth segmentation. More...
class	SurfaceDistancesMetric
	Compute a set of distance-based metrics between a label map and a ground truth segmentation. More...
class	PixelwiseClassificationMetrics
	Compute dense classification related metrics between a label map and a ground truth segmentation Precisely, computes the sensitivity, specificity, precision and recall. More...
struct	ElementTypeHash
	Hash function for the ElementType enum class. More...
class	TargetTag
	Simple data component to identify data that must be considered as learning target. More...
struct	Status
	Convenience tool for returning error messages together with statuses. More...
class	ModalitySynthesisAlgorithm
	Class for running an algorithm that that given an image in one modality generates the corresponding image in another modality. (e.g. MR to CT). More...
class	SamplingConfiguration
class	ModelConfiguration
	Configuration class for MachineLearningModel parameters. More...
class	MRIBiasFieldCorrectionAlgorithm
	Algorithm to perform bias field correction using an implicitly trained neural network This algorithm addresses the problem of intensity inhomogeneities "bias fields" in magnetic resonance imaging (MRI). More...
class	NetworkTrainingFileGenerator
	Generator of a Caffe neural network architecture text file from a set of the training data and some user-defined parameters This class is called from the GenerateCaffeTrainingDataAlgorithm to automatically produce a sample network suitable for training. More...
class	FileNotFoundError
class	IOError
struct	OperationException
	Custom exception to be used by Operation. More...
class	Operation
	Class to define a pre-post processing operation on data items. More...
class	InvertibleOperation
	Base class for operations that support inversion. More...
class	OperationsSequence
	Class to execute a sequence of Operation, depending on the workflow phase (training, validation, etc.). More...
class	OperationsSequenceAlgorithm
	Algorithm for running operations sequence on data. More...
class	OrderedMap
	Class for an insertion ordered map in order to mimic a python dictionary in C++ This is a convenience class which provides a similar interface to std::map and std::unordered_map It should be only used if there are not many elements to store or performance is not critical Due to its simple implementation it performs lookup only in O(N). More...
class	ProcessingRecordComponent
	Data component for keeping track of the data's origin. More...
class	RandomOperation
	Abstract class for random operations that build upon another one In order to create a randomized version of a BaseOperation, you need to derive from RandomOperation<BaseOperation> and just implement the method randomizeOperation. More...
class	Tensor
	Implementation of a Tensor class; used in TensorDataElement to directly present data to a neural network. More...
class	TensorSet
	Class for managing sets of tensors, one for each frame in the set. More...
struct	ReferenceImageDataComponent
	Data component used to store a reference image. More...
class	ImageResampler
	Helper class to run resampling algorithms within operations The class holds two permanent algorithm instances (one for images, one for labels) for the sake of efficiency Images with more 4 channels are properly handled so that GPU execution is still possible. More...
class	InverseOperation
	Operation that inverts a specific operation by using the InversionComponent. More...
class	AddCenterBoxOperation
	Add an additional channel to the input image with a binary box at its center. More...
class	AddPixelwisePredictionChannelOperation
	Run an existing pixelwise model and add result to the input image as additional channels. More...
class	AddPositionChannelOperation
	Add additional channels with the position of the pixels. More...
class	AddRandomNoiseOperation
	Apply a pixelwise random noise to the image intensities. More...
class	AdjustShiftScaleOperation
	Apply a shift and scale to each channel of the input image. More...
class	ApplyTopDownFlagOperation
	Flip the input image if it has a topDown flag set to false. More...
class	ApproximateToHigherResolutionOperation
	Replicate the input image from the original reference image (in ReferenceImageDataComponent). More...
class	ArgMaxOperation
	Create a label map with the indices corresponding of the input channel with the highest value. More...
class	AxisFlipOperation
	Flip image content along specified set of axes. More...
class	AxisRotationOperation
	Rotate image around image axis with axis-specific rotation angles that are signed multiples of 90 degrees. More...
class	BakeDeformationOperation
	Deform an image with its attached Deformation and store the result into the returned output image. More...
class	BakePhotometricInterpretationOperation
	Bake the Photometric Interpretation into the intensities of the image. More...
class	BakeTransformationOperation
	Apply the rotation contained in the matrix of the input volume. More...
class	BlobsFromKeypointsOperation
	Transforms keypoints into an actual image (blob map with the same size of the image). More...
class	CheckDataOperation
	Checks if all input data match a set of expected conditions. More...
class	ClipOperation
	Clip the intensities to a minimum and maximum value. More...
class	ConvertSlicesToVolumeOperation
	Stacks a set of 2D images extracted along a specified axis into an actual 3D volume. More...
class	ConvertToGrayOperation
	Convert the input image to a single channel image by averaging all channels. More...
class	ConvertVolumeToSlicesOperation
	Unstacks a 3D volume to a set of 2D images extracted along one of the axes. More...
class	ConvolutionalCRFOperation
	Adapt segmentation map or raw output of model to image content. More...
class	CopyOperation
	Copies a set of fields of a data item. More...
class	CropAroundLabelMapOperation
	Selects the given label from a multi-label, and crops the image and labels around it. More...
class	AddDegradedLabelAsChannelOperation
	Operation to add a channel with blobs whose sign depends on the label. More...
class	CropOperation
	Crop input images and label maps with a given size and offset. More...
class	CutOutOperation
	Cut out input images and label maps with a given size, offset and fill values. More...
class	DeformationOperation
	Apply a deformation to the image using a specified control point grid and specified displacements. More...
class	EnsureExplicitMaskOperation
	Converts the existing mask of all input images into explicit masks. More...
class	EnsureOneToOneMatrixMappingOperation
	Ensures that it is possible to get/set the matrix of each frame of the input image set independently. More...
class	ExtractRandomSubsetOperation
	Extracts a random subset from a SharedImageSet. More...
class	ExtractSubsetOperation
	Extracts a subset from a SharedImageSet. More...
class	GammaCorrectionOperation
	Apply a gamma correction which changes the overall contrast. More...
class	GenerateRandomKeypointsOperation
	Generate uniformly distributed random keypoints in the image. More...
class	HighPassOperation
	Smooths the input image with a Gaussian kernel with half_kernel_size, then subtracts the smoothed image from the input, resulting in a reduction of low-frequency components. More...
class	ImageMathOperation
	Computes a specified formula involving images from the input dataitem. More...
class	ImageMattingOperation
	Refine edges of label-map based on the intensities of the input image. More...
class	InvertOperation
	Invert the intensities of the image. More...
class	KeypointsFromBlobsOperation
	Extracts keypoints from a blob image Requires an image called label keypoints_field_name (str): Field name of the output keypoints keypoint_extraction_mode (int): Extraction mode: 0: Max, 1: Mean, 2: Local Max. More...
class	KeepLargestComponentOperation
	Create a label map with the largest components above the specified threshold. More...
class	ForegroundGuidedLabelUpsamplingOperation
	Operation that generates a label map by upsampling or resampling a multi-class one-hot encoded image (such as a softmax model prediction) to the space of a binary image (such as a sigmoid model prediction). More...
class	LinearIntensityMappingOperation
	Apply a linear shift and scale to the image intensities. More...
class	MRIBiasFieldCorrectionOperation
	Operation to perform bias field correction using an implicitly trained neural network (see MRIBiasFieldCorrectionAlgorithm for more details and the parameters description). More...
class	MRIBiasFieldGenerationOperation
	Apply or generate a multiplicative intensity modulation field. More...
class	MakeFloatOperation
	Convert the input image to float with original values (internal shifts and scales are baked in). More...
class	TagDataElementOperation
	Operation for changing data element tags. More...
class	MarkAsTargetOperation
	Mark elements from the input data item as "target" which might affect the behavior of subsequent operations that rely on Operation::ProcessingPolicy or use other custom target-specific logic. More...
class	UnmarkAsTargetOperation
	Remove the target tag from the elements of the input data item. More...
class	MergeAsChannelsOperation
	Merge multiple images into one along the channel dimension. More...
class	MorphologicalFilterOperation
	Runs a morphological operation on the input. More...
class	ConcatenateNeighboringFramesToChannelsOperation
	This function iterates over each frame, augmenting the channel dimension by appending or adding information from neighboring frames from both sides. More...
class	NormalizeMADOperation
	Normalize the input image based on robust statistics. More...
class	NormalizeNormalOperation
	Normalize the input image so that it has a zero-mean and a unit-standard deviation. More...
class	NormalizePercentileOperation
	Normalize the input image based on its intensity distribution, in particular on a lower and upper percentile. More...
class	NormalizeUniformOperation
	Normalize the input image based on their minimum/maximum intensity so that the output image has a [min; max] range. More...
class	OneHotOperation
	Encode a single channel label image to a one-hot representation with multiple channels. More...
class	PadOperation
	Pad an image to a specific padding size in each dimension. More...
class	PadDimsOperation
	Pad an image to specific target dimensions. More...
class	PadDimsToNextMultipleOperation
	Pads each dimension of the input image to the next multiple of the specified divisor. More...
class	PolyCropOperation
	Masks the image with a convex polygon as described in Markova et al. More...
class	RandomAddDegradedLabelAsChannelOperation
	Operation to add a channel with randomly distributed blobs, whose sign is positive if a blob is in the label and else negative. More...
class	RandomAddRandomNoiseOperation
	Apply AddRandomNoiseOperation to images with randomized intensity parameter. More...
class	RandomAxisFlipOperation
	Flip image content along specified set of axes, with independent sampling for each axis. More...
class	RandomAxisRotationOperation
	Rotate image around image axis with independently drawn axis-specific random rotation angle of +-{90, 180, 270} degrees. More...
class	RandomChoiceOperation
	Meta-operation that picks one operation from its configuration randomly and executes it This is particularly useful for image samplers, where we might want to alternate between different ways of sampling the input images The operationWeights argument allows to parameterize the operation selection probability distribution, if not specified, uniform sampling is used. More...
class	RandomCropAroundLabelMapOperation
	Random version of CropAroundLabelMapOperation that selects a single random label value and crops around it. More...
class	RandomCropOperation
	Crop input images and label maps with a matching random size and offset. More...
class	RandomCutOutOperation
	Apply a random cutout to the image. More...
class	RandomDeformationOperation
	Apply a deformation to the image using a specified control point grid and random displacements. More...
class	RandomGammaCorrectionOperation
	Apply a random gamma correction to the image intensities. More...
class	RandomImageFromLabelOperation
	Creates a random image from a label map, each label is sampled from a Gaussian distribution. More...
class	RandomInvertOperation
	Operation that randomly inverts an image with a default probability of 50% (can be changed). More...
class	RandomKeypointJitterOperation
	Adds an individually and randomly sampled offset to each keypoint of each KeypointElement. More...
class	RandomLinearIntensityMappingOperation
	Apply a random linear shift and scale to the image intensities. More...
class	RandomMRIBiasFieldGenerationOperation
	Apply or generate a random multiplicative intensity modulation field. More...
class	RandomPolyCropOperation
	Masks the image with a random convex polygon as described in Markova et al. More...
class	RandomResolutionReductionOperation
	Downsamples the image to a target_spacing and upsamples again to the original spacing to reduce image information. More...
class	RandomRotationOperation
	Rotate input images and label maps with random angles. More...
class	RandomScalingOperation
	Scale input images and label maps with random factors. More...
class	RandomSmoothOperation
	Apply a random smoothing on the image (Gaussian kernel). More...
class	RandomTemplateInpaintingOperation
	Inpaints a template into an image with randomly selected spatial and intensity transformation in a given range. More...
class	RecombinePatchesOperation
	Operation to recombine image patches back into a full image. More...
class	RectifyRotationOperation
	Sets the image matrix to the closest xyz-axis aligned rotation, effectively making every rotation angle a multiple of 90 degrees. More...
class	RemoveOperation
	Removes a set of fields from a data item. More...
class	RemoveMaskOperation
	Removes the mask of all input images. More...
class	RenameOperation
	Renames a set of fields of a data item. More...
class	ReplaceLabelsValuesOperation
	Replace some label values with other values (only works for integer-typed labels). More...
class	ResampleDimsOperation
	Resample the input to fixed target dimensions. More...
class	ResampleKeepingAspectRatioOperation
	Resample input to target dimensions while keeping aspect ratio of original images. More...
class	ResampleOperation
	Resample the input to a fixed target resolution. More...
class	ResampleToInputOperation
	Resample the input image with respect to the image in ReferenceImageDataComponent. More...
class	ResolutionReductionOperation
	Downsamples the image to the target_spacing and upsamples again to the original spacing to reduce image information. More...
class	RotationOperation
	Rotate input images and label maps with fixed angles. More...
class	RunModelOperation
	Run a machine learning model on the input item and merge the prediction to the input item. More...
class	ScalingOperation
	Scale input images and label maps with fixed factors. More...
class	SelectChannelsOperation
	Keeps a subset of the input channels specified by the selected channel indices (0-based indexing). More...
class	SetModalityOperation
	Sets the input modality. More...
class	SetLabelModalityOperation
	Sets the input modality. More...
class	SetMatrixToIdentityOperation
	Set the matrices of all images to identity (associated landmarks and boxes will be moved accordingly). More...
class	SetSpacingOperation
	Modify images so that image elements have specified spacing (associated landmarks and boxes will be moved accordingly). More...
class	SigmoidOperation
	Apply a sigmoid function on the input image. More...
class	SmoothOperation
	Run a convolution with a Gaussian kernel on the input image. More...
class	SoftmaxOperation
	Computes channel-wise softmax on input. More...
class	SplitIntoPatchesOperation
	Operation which splits the input image into overlapping patches for sliding window inference. More...
class	StandardizeImageAxesOperation
	Reorganize the memory buffer of a medical image to ensure anatomical consistency. More...
class	SwapImageAndLabelsOperation
	Swaps image and label map. More...
class	SyncOperation
	Synchronizes shared memory (CPU <-> OpenGL) of images. More...
class	TanhOperation
	Apply a tanh function on the input image. More...
class	TemplateInpaintingOperation
	Inpaints a template into an image with specified spatial and intensity transformation. More...
class	ThresholdOperation
	Threshold the input image to a binary map with only 0 or 1 values. More...
class	UndoPaddingOperation
	Apply the inverse of a previously applied padding operation. More...
class	CenterROISampler
	Sampler which samples one ROI from the input image and label map with a target size. More...
class	DefaultROISampler
	Sampler which simply returns the image and the label map, after padding of a specified dimension divisor: each spatial dimension of the output arrays will be divisible by dimensionDivisor. More...
class	LabelROISampler
	Sampler which samples ROIs from the input image and label map, such that one particular label appears. More...
class	OrientedROISampler
	The OrientedROISampler samples m_numSamples samples of size m_roiSize from each dataset with an efficient GPU sampler. More...
class	RandomROISampler
	Sampler which randomly samples ROIs from the input image and label map with a target The images will be padded if the target size is larger than the input image. More...
class	SplitROISampler
	Sampler which splits the input image into overlapping ROIs for sliding window inference. More...
class	BatchDataLoader
	Batches the items in the nested loader into batches of the specified size. More...
class	CacheDataLoader
	Caches the dataset loaded until now. More...
class	FilterDataLoader
	Filters the items according to user defined criterion. More...
class	FilterEmptyElementContentDataLoader
	Filters the items out if any of the selected fields is holding a DataElement with empty content. More...
class	InterleaveDataLoader
	Routes DataItems from different pipelines into one active pipeline. More...
class	MapDataLoader
	Applies a mapping to each item in the nested loader. More...
class	PersistentCacheDataLoader
	Caches the dataset loaded until now in a persistent manner (on a disk location). More...
class	PrefetchDataLoader
	Prefetch the next items in a background thread. More...
class	PreprocessDataLoader
	Applies a preprocessing pipeline to each item in the nested loader. More...
class	RepeatDataLoader
	Repeats items coming from a nested loader. More...
class	RandomizeDataLoader
	Maintains a fixed-size buffer that provides randomized items through iterative replacement. More...
class	SampleDataLoader
	Extract samples from each item in the nested loader. More...
class	ShuffleDataLoader
	Shuffles the next specified items in the nested loader. More...
class	SplitDataLoader
	Splits the items in the nested loader into DataItems containing a single image (batch size of 1). More...
class	BoundingBoxElement
	DataElement based on BoundingBoxSet. More...
class	ImageElement
	DataElement for 2D and 3D images. More...
class	KeypointsElement
	DataElement based on KeypointSet. More...
class	SISBasedElement
	Interface for DataElement based on SharedImageSet. More...
class	TensorSetElement
	DataElements for Raw Tensors (Experimental) Unlike ImageElements, these elements will go into the ML engine in the shape that they are created (i.e. More...
class	VectorElement
	DataElement for non-image data (such as imagewise labels), represented as 1D images (i.e. More...
struct	FileReaderColumn
	This struct represents a "column" of filenames in a filelist, like those contained in datalist_training.txt/data_list_validation.txt produced by ImFusionLabels. More...
class	FileReader
	Loads data by reading the files from a list of data "columns" (see above). More...
class	DataListReader
	DataReader to directly read data from a given datalist where each column corresponds to a (single) field of the DataItem. More...
class	InversionComponent
	Data component for storing operation inversion information. More...
class	PaddingDoneInfo
	Struct for geometrical information about how the patches were extracted from the original images. More...
class	PaddingDoneDataComponent
	Data component for keeping track of the original location of a patch in the original image. More...
struct	PatchInfo
	Struct for storing the descriptor of the image a patch was extracted from and the region of interest in the original image. More...
class	PatchesFromImageDataComponent
	Data component for keeping track of the original location of a patch in the original image. More...
class	PixelwiseLearningStream
	Stream running a pixelwise learning model on another stream. More...
class	MeshSegmentationAlgorithm
	Mesh segmentation algorithm based on a graph neural networks Converts an input Mesh to a Graph and runs a provided GNN model. More...
class	LabelsProjectDataReader
	Data loader for training pipelines that load data directly from a Labels project. More...

Typedefs
template<typename ParamType>
using	LoaderParam = ML::AdvancedParameter<ParamType, DataLoader>
using	DataLoaderList = std::vector<std::unique_ptr<DataLoader>>
using	DataLoaderDecoratorFactory = ML::Singleton<ML::Factory<DataLoader, DataLoaderList, const Properties&>>
using	DataReaderFactory = Singleton<Factory<DataReader, const Properties&>>
using	EngineFactory = Singleton<Registry<Factory<Engine, const Properties&>>>
	Engine Factory, this is instantiated in GenericFactory.cpp.
using	EngineSubFactory = Factory<Engine, const Properties&>
	Sub-factories (e.g. C++, Python) of the main factory.
template<typename ParamType>
using	EngineParameter = AdvancedParameter<ParamType, EngineConfiguration>
using	MetricFactory = Singleton<Factory<Metric, const Properties&>>
	Machine Learning Operation registry typedef.
template<typename ParamType>
using	SamplingParameter = AdvancedParameter<ParamType, SamplingConfiguration>
template<typename ParamType>
using	ModelParameter = AdvancedParameter<ParamType, ModelConfiguration>
template<typename ParamType>
using	OpParam = ML::AdvancedParameter<ParamType, Operation>
using	OperationFactory = Singleton<Registry<Factory<Operation, std::optional<const Properties>>>>
	Machine Learning Operation registry typedef.
using	OperationSubFactory = Factory<Operation, std::optional<const Properties>>
	Sub-factories (e.g. C++, Python) for the main factory.
using	FilterFunction = std::function<bool(const DataItem&)>
using	FilterFunctionRegistry = Singleton<Registry<FilterFunction>>
using	MapFunction = std::function<void(DataItem&)>
using	MapFunctionRegistry = Singleton<Registry<MapFunction>>
using	KeyMapping = std::unordered_map<size_t, DataItem::Field>
	Maps the indices in a DataList to field names to be used in the data item Note: the FileReader uses ML::open to load the content of a file, so in order to fill a DataItem with the data in the datalist we need to assign them to a field specified by the user.

Enumerations
enum class	ParamRequired { Yes , No }
	Enum to specify whether the parameter is required when configuring the parent class via the configure method.
enum class	ParamUnit { MM = 0 , Fraction = 1 , Voxel = 2 }
	Enum for the different types of parameter units. More...
enum class	Cardinality { Fixed = 0 , Dynamic = 1 , Infinite = 2 }
	Specifies the type of cardinality (or length) of a Dataset or a DataLoader, in particular whether the amount of produced items is fixed. More...
enum class	ExecutionProvider {   CPU = 0 , Custom = 1 , CUDA = 2 , DirectML = 3 ,   OpenVino = 4 , MPS = 5 }
	Options for acceleration providers used by engines.
enum class	ComputingDevice { ForceCPU = 0 , GPUIfGlImage , GPUIfOpenGl , ForceGPU }
	Computing device strategy. More...
enum class	Phase { Training = 1 , Validation = 2 , Inference = 4 , Always = Training | Validation | Inference }
	Various phases of a ML model - they can be used to control if some processing operations must be executed. More...
enum class	ModelType { Unknown = -1 , RandomForest , NeuralNetwork }
	Enum for different learning models. More...
enum class	PredictionType { Unknown = -1 , Classification , Regression , ObjectDetection }
	Types of prediction type. More...
enum class	PredictionOutput {   Unknown = -1 , Vector , Image , Keypoints ,   BoundingBoxes , Tensor }
	Types of prediction output. More...
enum class	ElementType {   Image = 0 , BoundingBox = 1 , Keypoint = 2 , Vector = 3 ,   Tensor = 4 }
	Types of elements that a data item might contain. More...
enum class	InterleaveMode { Alternate , Proportional , Unknown }
	Enum used to determine the behaviour of the InterleaveDataLoader. More...
enum class	ResetCriterion { Fixed , SmallestLoader , LargestLoader , Unknown }
	Reset criterion determines when interleaving dataloaders are reset. More...
enum class	RecombineMode { Default = 0 , Weighted }
	Recombination mode for handling overlapping patches when reconstructing an image from patches. More...

Functions
std::string	paramUnitToString (ParamUnit unit)
ParamUnit	stringToParamUnit (const std::string &unitStr)
bool	isCudaComputeCapabilitySupported (int deviceCCmajor, int deviceCCminor, int cudaRuntimeMajor, int cudaRuntimeMinor, const std::vector< CudaComputeCapability > &supportedComputeCapabilities)
	Check if the device is supported by the CUDA runtime.
Cardinality	operator&& (Cardinality first, Cardinality second)
void	debugMermaidOutput (DataLoader const &dataLoader, std::ostream &out)
	Output a Mermaid compatible flow graph to visualize the data flow.
std::optional< std::string >	isPythonOperationUsed (const std::vector< std::string > &opNames)
	Utility function for checking whether a python operation is used.
std::vector< DataLoaderSpecs >	propertyToDataLoaderSpecs (const Properties &properties)
std::vector< DataLoaderSpecs >	propertyListToDataLoaderSpecs (const PropertyList &propertyList)
std::vector< std::unique_ptr< Properties > >	dataLoaderSpecsToPropertyList (const std::vector< DataLoaderSpecs > &specs)
EngineSubFactory *	getSubEngineFactory (const std::string &subFactoryName)
	Get (and create if it does not exist yet) a sub-factory from a name.
EngineSubFactory *	getCppEngineFactory ()
	Helper function to get the C++ factory from the registry.
bool	isNotEmpty (DataItem const &dataItem)
	Returns false if the input data item is empty.
bool	hasNonZeroInputImage (DataItem const &dataItem)
	Returns false if any non-target SharedImageSets in the dataitem are images with only zeros in them, otherwise true.
bool	hasNonZeroTargetImage (DataItem const &dataItem)
	Returns false if any target SharedImageSets in the dataitem are images with only zeros in them, otherwise true.
template<typename MetricType>
std::unique_ptr< MetricType >	createMetric (const Properties &p)
OwningDataList	open (const std::string &filename)
	Similar to ApplicationController::open but does not require an ApplicationController.
std::optional< KeypointSet >	loadKeypoints (const std::string &path)
	Helper methods to load keypoints and bounding boxes.
std::optional< BoundingBoxSet >	loadBoundingBoxes (const std::string &path)
bool	saveKeypoints (const KeypointSet &keypoints, const std::string &path)
bool	saveBoundingBoxes (const BoundingBoxSet &bboxes, const std::string &path)
bool	landmarksToJson (const std::string &path, const KeypointSet &lms)
	Helper methods to load keypoints and bounding boxes from json files this allows to load keypoints and bboxes exported from ImFusionLabels as KeypointSet and BoundingBoxes.
bool	boundingBoxesToJson (const std::string &path, const BoundingBoxSet &bbs)
std::optional< KeypointSet >	landmarksFromJson (const std::string &path)
std::optional< BoundingBoxSet >	boundingBoxesFromJson (const std::string &path)
OperationSubFactory *	getSubOperationFactory (const std::string &subFactoryName)
	Get (and create if it does not exist yet) a sub-factory from a name.
OperationSubFactory *	getCppOperationFactory ()
	Helper function to get the C++ factory from the registry.
template<typename Op>
std::unique_ptr< Op >	makeUniqueOperation (std::optional< const Properties > p)
template<typename Op>
std::shared_ptr< Op >	makeOperation (std::optional< const Properties > p)
void	recordOperationForInversion (Operation &op, DataElement *element)
std::vector< Operation::Specs >	propertyToProcessingSpecs (const Properties &properties)
std::vector< Operation::Specs >	propertyListToProcessingSpecs (const PropertyList &propertyList)
std::vector< std::unique_ptr< Properties > >	processingSpecsToPropertyList (const std::vector< Operation::Specs > &specs)
bool	isTarget (const SharedImageSet &input)
	Check if an image is marked as a target.
void	tagAsTarget (SharedImageSet &input)
	Tag an image as a target.
void	untagAsTarget (SharedImageSet &input)
	Untag an image as label.
bool	isSemanticSegmentationMap (const SharedImageSet &input)
	Check if the image is a segmentation label map (with discrete values). This function should be used to determine how algorithms like downsampling/interpolation/etc. should treat this image, e.g. use nearest neighbor instead of linear interpolation.
ML::ComputingDevice	stringToDevice (const std::string &s)
	Convert string input to a computing device.
std::string	deviceToString (ML::ComputingDevice device)
	Convert computing device to string.
Phase	stringToPhase (const std::string &s)
	Convert string input to a phase.
std::string	phaseToString (Phase phase)
	Convert phase to a string.
std::string	modelTypeToString (ML::ModelType type)
	Convert prediction type to a string.
ML::ModelType	stringToModelType (const std::string &type)
	Convert prediction type to a string.
std::string	predictionTypeToString (ML::PredictionType type)
	Convert prediction type to a string.
ML::PredictionType	stringToPredictionType (const std::string &type)
	Convert prediction type to a string.
std::string	predictionOutputToString (ML::PredictionOutput type)
	Convert prediction output to a string.
ML::PredictionOutput	stringToPredictionOutput (const std::string &type)
	Convert string to prediction output.
std::string	elementTypeToString (ML::ElementType type)
	Convert ElementType to a string.
ML::ElementType	stringToElementType (const std::string &type)
	Convert string to ElementType.
std::string	interleaveModeToString (ML::InterleaveMode mode)
	Convert InterleaveMode to a string representation.
ML::InterleaveMode	stringToInterleaveMode (const std::string &type)
	Convert string representation to a InterleaveMode.
std::string	resetCriterionToString (ML::ResetCriterion mode)
	Convert ResetCriterion to a string representation.
ML::ResetCriterion	stringToResetCriterion (const std::string &type)
	Convert string representation to a ResetCriterion.
bool	shouldRunOnGPU (ML::ComputingDevice deviceStrategy, const SharedImageSet *image, bool allowMoreThan4Channels=false, std::function< void(std::string &&)> &&warnFunc=[](std::string &&) {})
	Return whether an algorithm should choose the GPU implementation.
std::map< std::string, std::vector< std::string > >	readDataList (std::string path, std::vector< std::string > prependKeys={"dataPath", "labelPath"})
	Read a data list file from an exported Labels project.
std::vector< std::vector< float > >	convertSharedImageSetToVectors (const SharedImageSet &input)
	Convert a shared image set to a vector of vector (useful for imagewise predictions).
std::string	modelPathToAbsoluteIfNotResource (const std::string &path, const std::string &refConfigFile="")
	Convert a model path to an absolute path, taking into account multiple special behaviour (sanitize path obtained by copy/pasting and "copying as path" from Windows explorer, i.e.
std::string	recombineModeToString (RecombineMode mode)
	Convert RecombineMode to a string.
RecombineMode	stringToRecombineMode (const std::string &str)
	Convert a string to RecombineMode.
bool	operator== (const PaddingDoneInfo &lhs, const PaddingDoneInfo &rhs)
bool	operator!= (const PaddingDoneInfo &lhs, const PaddingDoneInfo &rhs)
bool	operator== (const PatchInfo &lhs, const PatchInfo &rhs)
bool	operator!= (const PatchInfo &lhs, const PatchInfo &rhs)

Typedef Documentation

KeyMapping

using ImFusion::ML::KeyMapping = std::unordered_map<size_t, DataItem::Field>

Maps the indices in a DataList to field names to be used in the data item Note: the FileReader uses ML::open to load the content of a file, so in order to fill a DataItem with the data in the datalist we need to assign them to a field specified by the user.

Example1 : load(myfile.txt) -> DataList{sis1, sis2, sis3} a possible key mapping could be {{0, "data"}, {1, "label"}, {2, "mask"}}. The resulting DataItem will be: {{"data", sis1}, {"label", sis2}, {"mask", sis3}}

If you don't need all the data in the datalist, you can skip some by not specifying an index mapping for the data you want to skip

Example 2: load(myfile.txt) -> DataList{sis1, sis2, sis3} mapping: {{0, "data"}, {2, "mask"}}. The resulting DataItem will be: {{"data", sis1}, {"mask", sis3}}

Enumeration Type Documentation

ParamUnit

enum class ImFusion::ML::ParamUnit

strong

Enum for the different types of parameter units.

Enumerator
MM	parameter provided in mm
Fraction	parameter provided in fractional units of the image extent
Voxel	parameter provided in voxel units

Cardinality

enum class ImFusion::ML::Cardinality

strong

Specifies the type of cardinality (or length) of a Dataset or a DataLoader, in particular whether the amount of produced items is fixed.

For instance, a data loader that splits a set of images into individual images cannot know in advance the number of items to be produced since it depends on the actual content of the loaded files.

Enumerator
Fixed	The number of items to be produced can be determined in advance (i.e. without depending on the item content).
Dynamic	The number of items to be produced cannot be determined in advance but is guaranteed to be finite.
Infinite	The number of items to be produced is infinite.

ComputingDevice

enum class ImFusion::ML::ComputingDevice

strong

Computing device strategy.

Enumerator
ForceCPU	Force computation on the CPU.
GPUIfGlImage	Use GPU only if the image has already been uploaded to the GPU.
GPUIfOpenGl	Use GPU only if an OpenGL context is available.
ForceGPU	Force computation on the GPU (throws if not possible).

Phase

enum class ImFusion::ML::Phase

strong

Various phases of a ML model - they can be used to control if some processing operations must be executed.

For instance, input normalization should always be executed, but random augmentations would typically only be executed at training time.

Enumerator
Training	Training of the model, i.e. when its parameter are optimized.
Validation	Evaluation of the model during training (only to compute the metrics on a held out dataset).
Inference	When the model is actually deployed and ran on a new input.
Always	Convenience value to include all phases.

ModelType

enum class ImFusion::ML::ModelType

strong

Enum for different learning models.

Enumerator
Unknown	Default, invalid value.
RandomForest	Random forest (not supported anymore).
NeuralNetwork	Neural network (standard value).

PredictionType

enum class ImFusion::ML::PredictionType

strong

Types of prediction type.

Enumerator
Unknown	Default, invalid value.
Classification	Prediction is categorical.
Regression	Prediction is a continuous value.
ObjectDetection	Prediction is a set of BoundingBox, plus optionally Keypoints and Masks.

PredictionOutput

enum class ImFusion::ML::PredictionOutput

strong

Types of prediction output.

Enumerator
Unknown	Default, invalid value.
Vector	Output size does not depend on the input size.
Image	Output has the same size as the input.
Keypoints	Output is a set of keypoints.
BoundingBoxes	Output is a set of bounding boxes.
Tensor	Output is a set of tensors.

ElementType

enum class ImFusion::ML::ElementType

strong

Types of elements that a data item might contain.

Enumerator
Image	Element representing a set of 2D or 3D images.
BoundingBox	Element representing a set of 2D or 3D bounding boxes.
Keypoint	Element representing a set of 2D or 3D landmarks.
Vector	Element representing a set of scalar (1D) values.
Tensor	Element representing an arbitrary array of values, for instance used for graphs.

InterleaveMode

enum class ImFusion::ML::InterleaveMode

strong

Enum used to determine the behaviour of the InterleaveDataLoader.

Enumerator
Alternate	Continuously interleaves each dataloader one dataitem at a time regardless of size until the reset criterion is met.
Proportional	Interleaves dataloaders proportional to their size.
Unknown	Default, invalid value.

ResetCriterion

enum class ImFusion::ML::ResetCriterion

strong

Reset criterion determines when interleaving dataloaders are reset.

Enumerator
Fixed	Use a fixed parameter to control the epoch size.
SmallestLoader	Stop as soon as the smallest loader is exhausted.
LargestLoader	Stop when the largest loader is exhausted.
Unknown	Default, invalid value.

RecombineMode

enum class ImFusion::ML::RecombineMode

strong

Recombination mode for handling overlapping patches when reconstructing an image from patches.

This determines how pixel values from overlapping patches are combined. See RecombinePatchesOperation for more details on each mode.

Enumerator
Default	Simple averaging of patch voxels in overlapping regions.
Weighted	Each patch is decayed at its borders with a Gaussian function.

Function Documentation

isCudaComputeCapabilitySupported()

bool ImFusion::ML::isCudaComputeCapabilitySupported	(	int	deviceCCmajor,
		int	deviceCCminor,
		int	cudaRuntimeMajor,
		int	cudaRuntimeMinor,
		const std::vector< CudaComputeCapability > &	supportedComputeCapabilities )

Check if the device is supported by the CUDA runtime.

Parameters

deviceCCmajor	The major version of the GPU
deviceCCminor	The minor version of the GPU
cudaRuntimeMajor	The major version of the CUDA runtime
cudaRuntimeMinor	The minor version of the CUDA runtime
supportedComputeCapabilities	The list of supported Compute Capabilities of the CUDA library

open()

OwningDataList ImFusion::ML::open

(

const std::string &

filename

)

Similar to ApplicationController::open but does not require an ApplicationController.

Cannot open all files that ApplicationController supports (e.g. workspaces).

isTarget()

bool ImFusion::ML::isTarget

(

const SharedImageSet &

input

)

Check if an image is marked as a target.

This function should be used to determine whether it is an input data or a target data. Note: if an image is tagged with deprecated LabelTag, this function consideres it a target and returns true.

untagAsTarget()

void ImFusion::ML::untagAsTarget

(

SharedImageSet &

input

)

Untag an image as label.

Note: if an image is tagged with deprecated LabelTag, this function removes the LabelTag as well

readDataList()

std::map< std::string, std::vector< std::string > > ImFusion::ML::readDataList	(	std::string	path,
		std::vector< std::string >	prependKeys = {"dataPath", "labelPath"} )

Read a data list file from an exported Labels project.

Returns a string to vector map, where the keys are defined by the header of the file.

modelPathToAbsoluteIfNotResource()

std::string ImFusion::ML::modelPathToAbsoluteIfNotResource	(	const std::string &	path,
		const std::string &	refConfigFile = "" )

Convert a model path to an absolute path, taking into account multiple special behaviour (sanitize path obtained by copy/pasting and "copying as path" from Windows explorer, i.e.

remove file:/// prefix and double quotes, or prepending the ML default model path). When calling this function on a sidecar file, the original path to the configuration file can be provided so that the sidecar file can be searched in the same folder.