Digikam::DNNFaceDetectorBase Class Reference

Inheritance diagram for Digikam::DNNFaceDetectorBase:

Public Member Functions
	DNNFaceDetectorBase (float scale, const cv::Scalar &val, const cv::Size &inputImgSize)
virtual void	detectFaces (const cv::Mat &inputImage, const cv::Size &paddedSize, std::vector< cv::Rect > &detectedBboxes)=0
cv::Size	nnInputSizeRequired () const
virtual void	setFaceDetectionSize (FaceScanSettings::FaceDetectionSize faceSize)

Static Public Attributes
static float	nmsThreshold = 0.4F
	Threshold for nms suppression.
static int	uiConfidenceThreshold = DNN_MODEL_THRESHOLD_NOT_SET
	Threshold for bbox detection. It can be init and changed in the GUI.

Protected Member Functions
void	correctBbox (cv::Rect &bbox, const cv::Size &paddedSize) const
void	selectBbox (const cv::Size &paddedSize, float confidence, int left, int right, int top, int bottom, std::vector< float > &goodConfidences, std::vector< cv::Rect > &goodBoxes, std::vector< float > &doubtConfidences, std::vector< cv::Rect > &doubtBoxes) const

Protected Attributes
cv::Size	inputImageSize = cv::Size(300, 300)
cv::Scalar	meanValToSubtract = cv::Scalar(0.0, 0.0, 0.0)
DNNModelBase *	model = nullptr
float	scaleFactor = 1.0F

Member Function Documentation

selectBbox()

void Digikam::DNNFaceDetectorBase::selectBbox ( const cv::Size & paddedSize, float confidence, int left, int right, int top, int bottom, std::vector< float > & goodConfidences, std::vector< cv::Rect > & goodBoxes, std::vector< float > & doubtConfidences, std::vector< cv::Rect > & doubtBoxes ) const

protected

Classify bounding boxes detected. Good bounding boxes are defined as boxes that reside within the non-padded zone or those that are out only for min of (10% of padded range, 10% of bbox dim).

Bad bounding boxes are defined as boxes that have at maximum 25% of each dimension out of non-padded zone.