DeepLearning_ClassifyObject

Header:	AVLDL.h
Namespace:	avl

Performs whole image classification using a trained deep-learning model.

Syntax

void avl::DeepLearning_ClassifyObject
(
	const avl::Image& inImage,
	atl::Optional<const avl::Rectangle2D&> inRoi,
	atl::Optional<const avl::CoordinateSystem2D&> inRoiAlignment,
	const avl::DeepModel_Classification& inDeepModel,
	const bool inCreateRelevanceHeatmap,
	atl::Array<avl::ClassConfidence>& outConfidences,
	atl::String& outClassName,
	int& outClassIndex,
	float& outScore,
	avl::Image& outRelevanceHeatmap
)

Parameters

Name	Type	Default	Description
inImage	const Image&		Input image
inRoi	Optional<const Rectangle2D&>	NIL	Area of interest
inRoiAlignment	Optional<const CoordinateSystem2D&>	NIL
inDeepModel	const DeepModel_Classification&		Trained model
inCreateRelevanceHeatmap	const bool	False	Enable-Disable create relevance heatmap
outConfidences	Array<ClassConfidence>&		Returns confidences for all classes
outClassName	String&		Returns name of the class with highest confidence
outClassIndex	int&		Returns the index of the class with highest confidence
outScore	float&		Returns the value of the highest confidence
outRelevanceHeatmap	Image&		Returns heatmap indicating how strong specific parts of image influenced classification result

Requirements

For input inImage only pixel formats are supported: 1⨯uint8, 3⨯uint8.

Read more about pixel formats in Image documentation.

Hints

Loading Deep Learning Model may take longer timer. Consider using DeepLearning_LoadModel_ObjectClassification for pre-loading model before execution starts.

Remarks

Model provided on inDeepModel input will be loaded automatically on first usage of this filter.
This filter should not be executed along with running Deep Learning Service as it may result in degraded performance or even out-of-memory errors. However, if stopping Deep Learning Service is not an option (e.g. program uses other Deep Learning filters related to Anomaly Detection or Instances Segmentation) it is advised that this filter should be executed before other Deep Learning filters. Such "warm up" execution does not have to use real image or roi as long as provided image and roi have similar size.

Multithreaded environment

This function is not guaranteed to be thread-safe. When used in multithreaded environment, it has to be manually synchronized.

Errors

List of possible exceptions:

Error type	Description
DomainError	Not supported inImage pixel format in DeepLearning_ClassifyObject.