ModelFactory Class Reference

#include <AnatomyPlugin/include/ImFusion/AnatomyPlugin/ModelFactory.h>

Factory for managing different versions of ML model sets globally. More...

Detailed Description

Factory for managing different versions of ML model sets globally.

Such a model set is a structured collection of (e.g. per-anatomy) ML models. Can be used for customers to override default models in anatomy-specific segmentation and registration algorithms.

Example usages:

Implement modelFactoryInstance() to share a global collection of model sets across instances of a particular class:

static ModelFactory& modelFactoryInstance() {
    auto initModelFactory = [](){
        ModelFactory::ModelSet defaultModelSet(
            "ImFusion Standard",
            "2024.v1",
            {
                {
                    ModelIdentifier::KIDNEY_SEGMENTATION_MODEL,
                    { Keys::RightKidney::key, Keys::LeftKidney::key },
                    {
                        {ModelFactory::MLEngine::Torch, "<torchPath>"},
                        {ModelFactory::MLEngine::ONNX, "<onnxPath>"},
                        {ModelFactory::MLEngine::TensorRT, "<tensorRtPath>"},
                    }
                }
            }
        );
        defaultModelSet.metadata.setParam("Description", "This is a description.");
        defaultModelSet.metadata.setParam("KeypointThreshold", 0.4);
        auto modelFactory = std::make_unique<ModelFactory>();
        modelFactory->registerModelSet(defaultModelSet, true);
        return modelFactory;
    };
    static std::unique_ptr<ModelFactory> instance = initModelFactory();
    return *instance;
}

Create a customized model set from the default model set by copying and modifying it:

if (auto defaultModelSet = MyAlgorithmBase::modelFactoryInstance().defaultModelSet())
{
    // Copy the default ModelSet
    ModelFactory::ModelSet modelSet = *defaultModelSet;
    // ... modify the modelSet while keeping some models from the default model set and register it.
    // Note: Registering the new ModelSet without changing the name or version replaces the default ModelSet.
    MyAlgorithmBase::modelFactoryInstance().registerModelSet(modelSet);
}

Convenience function for getting the appropriate model path for the ML::Engine passed:

ModelInfo myModel = modelSet.modelIdentifierToModelInfo.at(ModelIdentifier::KIDNEY_SEGMENTATION_MODEL);
std::optional<std::string> modelPath = ModelFactory::ModelInfo::modelPath(myModel, ModelFactory::MLEngine::Torch);

Classes
struct	MLEngine
	Currently supported ML Engines (cf. ML::Engine) More...
struct	ModelInfo
	Holds the meta information about a single model. More...
struct	ModelSet
	List of models. More...

Public Types
using	ModelIdentifier = std::string
using	MLEngineName = std::string

Public Member Functions
	ModelFactory ()=default
	Create a model factory.
ModelFactory &	operator= (const ModelFactory &)=delete
ModelFactory &	operator= (ModelFactory &&)=default
	ModelFactory (ModelFactory &&)=default
ModelFactory	clone () const
	Create a copy of the current model factory, can be used to assign a model factory (via operator=(ModelFactory&&)) even though the usual assignment operator has been deleted.
void	registerModelSet (const ModelSet &modelSetIn, bool asNewDefault=false)
	Registers the given ModelSet in the factory.
std::vector< ModelSet::Name >	modelSetNames () const
	Returns the names of all registered ModelSets.
std::vector< ModelSet::Version >	modelSetVersions (const ModelSet::Name &modelSetName) const
	Returns the versions of all registered ModelSets with the given ModelSet::Name.
std::shared_ptr< const ModelSet >	modelSet (const ModelSet::Name &modelSetName, const ModelSet::Version &modelSetVersion) const
	Returns the ModelSet with the given ModelSet::Name and ModelSet::Version if it exists, nullptr otherwise.
std::shared_ptr< const ModelSet >	defaultModelSet () const
	Returns the default ModelSet. nullptr if no model set was registered as default yet.
std::shared_ptr< const ModelSet >	lastRegisteredModelSet () const
	Returns the last registered ModelSet. nullptr if no model set was registered yet.
bool	anyModelAvailable (const ModelIdentifier &modelIdentifier) const
	Returns true if any model set contains a model with the given identifier.
std::optional< std::string >	modelPath (const ModelSet::Name &modelSetName, const ModelSet::Version &modelSetVersion, const ModelIdentifier &modelIdentifier, const MLEngineName &engineName) const
	Returns true if the model set contains a model with the given identifier for the given engine.
bool	hasModel (const ModelSet::Name &modelSetName, const ModelSet::Version &modelSetVersion, const ModelIdentifier &modelIdentifier, const MLEngineName &engineName) const
	Returns true if the model set contains a model with the given identifier for the given engine.
std::unique_ptr< ML::MachineLearningModel >	createModel (const ModelSet::Name &modelSetName, const ModelSet::Version &modelSetVersion, const ModelIdentifier &modelIdentifier, const MLEngineName &engineName, bool verbose=true) const
	Convenience function to fetch model information from the factory and directly instantiate it using the ML plugin.

Member Function Documentation

registerModelSet()

void registerModelSet	(	const ModelSet &	modelSetIn,
		bool	asNewDefault = false )

Registers the given ModelSet in the factory.

A potentially existing ModelSet with the same ModelSet::Name and ModelSet::Version combination is overridden. To set the model set as the new default (cf. defaultModelSet()) pass true for asNewDefault.

createModel()

std::unique_ptr< ML::MachineLearningModel > createModel ( const ModelSet::Name & modelSetName, const ModelSet::Version & modelSetVersion, const ModelIdentifier & modelIdentifier, const MLEngineName & engineName, bool verbose = true ) const

nodiscard

Convenience function to fetch model information from the factory and directly instantiate it using the ML plugin.

Returns nullptr if the model is not stored in the ModelFactory or the ML engine is not configured correctly.

---

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

• AnatomyPlugin/include/ImFusion/AnatomyPlugin/ModelFactory.h