ImFusion C++ SDK 4.5.0
ImFusion::ML::ChannelGatherOperation Class Reference

#include <ImFusion/ML/Operations/ChannelGatherOperation.h>

Gathers one scalar per voxel from a multi-channel "value" image using per-voxel channel indices supplied in a separate "index" (label-map) image. More...

Inheritance diagram for ImFusion::ML::ChannelGatherOperation:

Detailed Description

Gathers one scalar per voxel from a multi-channel "value" image using per-voxel channel indices supplied in a separate "index" (label-map) image.

For every voxel v: output[v, k] = values[v, indices[v, k]] for k in [0, indexChannels)

The index image must be an integer type (uint8 or uint16) and may have one or more channels. If the index image has K channels, the output is a K-channel image (same type as the value input) where each channel is an independent gather. A single-channel index image (e.g. an ArgMax label map) applied to a softmax probability image produces a single-channel confidence map.

The value image may have any pixel type (float, double, uint8, int16, etc.) with C channels (C >= max(index value) + 1). The output has the same pixel type as the value input, preserving the ImageDescriptor (including shift and scale). Both images must have identical spatial dimensions (width x height x slices).

Index values that are negative or exceed the number of value channels cause an OperationException with a diagnostic message identifying the offending voxel and frame.

Only CPU execution is supported. Requesting GPU will throw.

The three DataItem field names are configurable via p_valueField, p_indexField, and p_outputField. The output replaces or creates the element at outputField; the value and index elements are left unchanged (unless outputField coincides with one of them).

Public Member Functions

 ChannelGatherOperation (std::string valueField="values", std::string indexField="indices", std::string outputField="gathered")
void process (DataItem &item) override
 Processes the (value, index) pair and writes the gathered result to the output field.
bool doesNotModifyInput () const override
 Returns whether the operation is guaranteed to not modify its input element, either because it computes its output out-of-place or because it is a no-op.
Public Member Functions inherited from ImFusion::ML::Operation
 Operation (std::string name, ProcessingPolicy processingPolicy)
 Derived classes must specify a name and whether, by default, the operation should also be applied to label maps.
virtual std::shared_ptr< SharedImageSetprocess (std::shared_ptr< SharedImageSet > input)
 Utility function to directly apply an operation on a SharedImageSet irrespective of active fields and processing policy.
virtual std::shared_ptr< KeypointSetprocess (std::shared_ptr< KeypointSet > input)
 Utility function to directly apply an operation on a set of KeypointSet irrespective of active fields and processing policy.
virtual std::shared_ptr< BoundingBoxSetprocess (std::shared_ptr< BoundingBoxSet > input)
 Utility function to directly apply an operation on a set of BoundingBoxSet irrespective of active fields and processing policy.
virtual std::unique_ptr< SharedImageSetprocess (std::unique_ptr< SharedImageSet > input) final
 Utility function with unique_ptr as input and output to avoid breaking changes (must not be overridden).
void addTemporaryPreProcessHook (PreProcessHook hook)
 Add a temporary hook that will only be active for the next process call and then automatically removed after the DataItem is processed.
void addTemporaryPostProcessHook (PostProcessHook hook)
 Add a temporary hook that will only be active for the next process call and then automatically removed after the DataItem is processed.
virtual bool checkRequiredItemsTypes (const DataItem &item) const
 Checks if the data item holds fields with types required by the operation.
const std::stringname () const
 Returns the name of the operation.
virtual bool configure (const Properties &properties)
 Configure an Operation with the given properties and return whether the configuration was successful.
virtual Properties configuration () const
 Return the Operation configuration.
void setActiveFields (std::optional< std::unordered_set< std::string > > activeFields)
std::optional< std::unordered_set< std::string > > activeFields () const
virtual void setProcessingPolicy (ProcessingPolicy policy)
ProcessingPolicy processingPolicy () const
std::optional< uint32_t > seed () const
 Return the current seed (if there is one).
virtual void seedRandomEngine (uint32_t seed)
 Set a seed for the random generator.
ML::ComputingDevice computingDevice () const
 Get the computing device.
virtual void setComputingDevice (ML::ComputingDevice device)
 Set the computing device selection strategy.
void configFailed (const std::string &missingParam) const
 Helper function to show an error message due to a bad configuration.
void logDeprecatedParam (const std::string &oldName, const std::string &newName) const
 Helper function to print a warning because of a deprecated parameter has been specified.
bool errorOnUnexpectedBehaviour () const
 Get the policy whether to treat unexpected behavior warnings as errors.
virtual void setErrorOnUnexpectedBehaviour (bool error)
 Treat unexpected behavior warnings as errors.
void registerParameter (ParameterBase *param)
 Register a parameter so that the operation knows about it, and can automatically configure it (unless it has been marked as manually configured parameter via the Operation::setManuallyConfiguredParameters method).
std::vector< ParameterBase * > parameters () const
 Return the list of registered parameters.
virtual bool supportsInversion () const
 Whether the operation supports inversion. Returns False unless the operation derives from InvertibleOperation.
const std::stringrecordIdentifier () const
 Get the operation's record identifier.
virtual void setRecordIdentifier (const std::string &recordIdentifier)
 Set the operation's record identifier.
const std::stringlogDomain () const
 Log domain for AdvancedParameter.
std::unordered_set< std::stringselectedFields (const DataItem &item) const
 Get the fields on which the operation will be applied.

Public Attributes

 ML_OPERATION_DISABLE_UTILITY_PROCESS
OpParam< std::stringp_valueField = {"value_field", std::string("values"), this, ParamRequired::No}
 DataItem field name for the multi-channel value image (e.g. softmax probabilities).
OpParam< std::stringp_indexField = {"index_field", std::string("indices"), this, ParamRequired::No}
 DataItem field name for the integer index image (e.g. label map from ArgMax).
OpParam< std::stringp_outputField = {"output_field", std::string("gathered"), this, ParamRequired::No}
 DataItem field name where the gathered output will be stored.

Additional Inherited Members

Public Types inherited from ImFusion::ML::Operation
enum  ProcessingPolicy { EverythingExceptLabels = 0 , Everything , OnlyLabels }
 Policy used by default when selecting on the fields on which the operation will be applied Note that "Labels" here refers to the target tag, not to be confused with Data::Modality::LABEL. More...
using PreProcessHook = std::function<void(Operation&, DataElement*)>
using PostProcessHook = std::function<void(Operation&, DataElement*)>
Static Public Member Functions inherited from ImFusion::ML::Operation
static std::string processingPolicyToString (const ProcessingPolicy policy)
static ProcessingPolicy stringToProcessingPolicy (const std::string &s)
static std::unique_ptr< OperationcreateFromFactories (const Operation::Specs &specs)
 Helper function to create an operation from any factory (both C++ and Python factories are tested).
Protected Member Functions inherited from ImFusion::ML::Operation
virtual std::shared_ptr< SharedImageSetprocessImages (std::shared_ptr< SharedImageSet > input) const
 Virtual function that every subclass should override.
virtual std::shared_ptr< BoundingBoxSetprocessBoxes (std::shared_ptr< BoundingBoxSet > input) const
 Virtual function that every subclass should override.
virtual std::shared_ptr< KeypointSetprocessPoints (std::shared_ptr< KeypointSet > input) const
 Virtual function that every subclass should override.
virtual std::shared_ptr< SharedImageSetprocessVectors (std::shared_ptr< SharedImageSet > input) const
 Virtual function that every subclass should override.
virtual std::shared_ptr< TensorSetprocessTensors (std::shared_ptr< TensorSet > input) const
 Virtual function that every subclass should override.
void throwOperationError (const std::string &msg) const
 Helper function to throw an exception due to a runtime error.
void warnOperationUnexpectedBehaviour (const std::string &msg) const
 Helper function to warn about a behavior different than asked.
bool inputIsEmptyOrNull (const SharedImageSet *input) const
 Helper function to check if the input is empty or null and print a warning if so.
virtual bool useGPU (const SharedImageSet *input) const
 Helper function that returns which device the operation should choose based on its configuration and a given input.
void prepareInputForDevice (SharedImageSet &input) const
 Make sure the input is on the correct device for the current device strategy.
virtual bool allowChannelBatchOnGPU () const
 Whether the operation implements channel-batching so that it can still be run on input with more than 4 channels.
bool configureOnly (const Properties &properties, const std::vector< ParameterBase * > &paramSelection) noexcept
 Helper method to configure only a subset of parameters.
void setManuallyConfiguredParameters (const std::vector< ParameterBase * > &manuallyConfiguredParams)
 Disable the auto-configuration of some of the parameters so that they can be parsed manually (in the configure function of the derived class).
void applyPreProcessHooks (DataElement *element)
 Apply the current pre-process hooks to an element.
void applyPostProcessHooks (DataElement *element)
 Apply the current post-process hooks to an element.
Protected Attributes inherited from ImFusion::ML::Operation
std::string m_name
ProcessingPolicy m_processingFieldsPolicy
 Flag specifying whether labels should be processed by default (when no active field has been specified).
std::optional< std::unordered_set< std::string > > m_activeFields
 Run the Operation only on those fields (if empty, will select suitable fields based on the current processing policy) when calling process().
std::unordered_set< std::stringm_alreadyWarned
 Set of warnings about unexpected behavior that have already been printed.
bool m_errorOnUnexpectedBehaviour = false
 Whether to throw an exception instead of warning about unexpected behavior.
std::optional< uint32_t > m_seed
Random::Generator m_randGenerator
ML::ComputingDevice m_device = ML::ComputingDevice::GPUIfOpenGl
std::vector< ParameterBase * > m_params
 All registered parameters.
std::vector< ParameterBase * > m_manuallyConfiguredParams
 All registered parameters that should not be configured automatically.
std::unordered_map< std::string, ElementTypem_requiredFieldsTypes
 List of all required field types used by checkRequiredItemsTypes.
std::string m_recordIdentifier
 User-provided unique identifier for this operation, used for recording processing history and inversion.
std::vector< PreProcessHook > m_preProcessHooks
std::vector< PostProcessHook > m_postProcessHooks
std::vector< PreProcessHook > m_temporaryPreProcessHooks
 Pre-process hooks that will be used for the next processing call only.
std::vector< PostProcessHook > m_temporaryPostProcessHooks
 Post-process hooks that will be used for the next processing call only.

Member Function Documentation

◆ process()

void ImFusion::ML::ChannelGatherOperation::process ( DataItem & item)
overridevirtual

Processes the (value, index) pair and writes the gathered result to the output field.

Reimplemented from ImFusion::ML::Operation.

◆ doesNotModifyInput()

bool ImFusion::ML::ChannelGatherOperation::doesNotModifyInput ( ) const
inlineoverridevirtual

Returns whether the operation is guaranteed to not modify its input element, either because it computes its output out-of-place or because it is a no-op.

This can be used to decide whether a copy of the input should be made before calling the Operation::process function. Note that this only relates to the overloads taking specific elements as inputs (e.g. images, keypoints, etc.), since the Operation::process(DataItem& item) function always modifies its input.

Reimplemented from ImFusion::ML::Operation.


The documentation for this class was generated from the following file:
  • ImFusion/ML/Operations/ChannelGatherOperation.h
Search Tab / S to search, Esc to close