The BBox class processes the NormalizedBBox data used with SSD. More...

Inheritance diagram for MyCaffe.common.BBoxUtility< T >:

Public Member Functions
	BBoxUtility (CudaDnn< T > cuda, Log log)
	The constructor. More...

void	Dispose ()
	Clean up all resources. More...

float	ComputeAP (List< Tuple< float, int > > rgTp, int nNumPos, List< Tuple< float, int > > rgFp, ApVersion apVersion, out List< float > rgPrec, out List< float > rgRec)
	Compute the average precision given true positive and false positive vectors. More...

List< int >	CumSum (List< Tuple< float, int > > rgPairs)
	Calculate the cumulative sum of a set of pairs. More...

void	ApplyNMSFast (List< NormalizedBBox > rgBBoxes, List< float > rgScores, float fScoreThreshold, float fNmsThreshold, float fEta, int nTopK, out List< int > rgIndices)
	Do a fast non maximum supression given bboxes and scores. More...

List< int >	ApplyNMS (List< NormalizedBBox > rgBBoxes, List< float > rgScores, float fThreshold, int nTopK)
	Do non maximum supression given bboxes and scores. More...

List< int >	ApplyNMS (List< NormalizedBBox > rgBBoxes, List< float > rgScores, float fThreshold, int nTopK, bool bReuseOverlaps, out Dictionary< int, Dictionary< int, float > > rgOverlaps)
	Do non maximum supression given bboxes and scores. More...

Dictionary< int, Dictionary< int, List< NormalizedBBox > > >	GetDetectionResults (float[] rgData, int nNumDet, int nBackgroundLabelId)
	Get detection results from rgData. More...

List< NormalizedBBox >	GetPrior (float[] rgPriorData, int nNumPriors, out List< List< float > > rgPriorVariances)
	Get the prior boundary boxes from the rgPriorData. More...

List< List< float > >	ComputeConfLoss (float[] rgConfData, int nNum, int nNumPredsPerClass, int nNumClasses, int nBackgroundLabelId, MultiBoxLossParameter.ConfLossType loss_type)
	Compute the confidence loss for each prior from rgConfData. More...

List< List< float > >	ComputeConfLoss (float[] rgConfData, int nNum, int nNumPredsPerClass, int nNumClasses, int nBackgroundLabelId, MultiBoxLossParameter.ConfLossType loss_type, List< DictionaryMap< List< int > > > rgAllMatchIndices, DictionaryMap< List< NormalizedBBox > > rgAllGtBoxes)
	Compute the confidence loss for each prior from rgConfData. More...

List< List< float > >	ComputeConfLoss (float[] rgConfData, int nNum, int nNumPredsPerClass, int nNumClasses, int nBackgroundLabelId, MultiBoxLossParameter.ConfLossType loss_type, List< Dictionary< int, List< int > > > rgAllMatchIndices, Dictionary< int, List< NormalizedBBox > > rgAllGtBoxes)
	Compute the confidence loss for each prior from rgConfData. More...

List< Dictionary< int, List< float > > >	GetConfidenceScores (float[] rgConfData, int nNum, int nNumPredsPerClass, int nNumClasses)
	Calculate the confidence predictions from rgConfData. More...

List< LabelBBox >	GetLocPredictions (float[] rgLocData, int nNum, int nNumPredsPerClass, int nNumLocClasses, bool bShareLocation)
	Create a set of local predictions from the rgLocData. More...

DictionaryMap< List< NormalizedBBox > >	GetGroundTruth (float[] rgGtData, int nNumGt, int nBackgroundLabelId, bool bUseDifficultGt)
	Create a set of ground truth bounding boxes from the rgGtData. More...

Dictionary< int, LabelBBox >	GetGroundTruthEx (float[] rgGtData, int nNumGt, int nBackgroundLabelId, bool bUseDifficultGt)
	Create a set of ground truth bounding boxes from the rgGtData. More...

void	Match (List< NormalizedBBox > rgGtBboxes, List< NormalizedBBox > rgPredBboxes, int nLabel, MultiBoxLossParameter.MatchType match_type, float fOverlapThreshold, bool bIgnoreCrossBoundaryBbox, out List< int > rgMatchIndices, out List< float > rgMatchOverlaps)
	Find matches between a list of two bounding boxes. More...

bool	IsCrossBoundary (NormalizedBBox bbox)
	Returns whether or not the bbox is overlaps outside the range [0,1] More...

List< LabelBBox >	DecodeAll (List< LabelBBox > rgAllLocPreds, List< NormalizedBBox > rgPriorBboxes, List< List< float > > rgrgfPrioVariances, int nNum, bool bShareLocation, int nNumLocClasses, int nBackgroundLabelId, PriorBoxParameter.CodeType codeType, bool bVarianceEncodedInTarget, bool bClip)
	Decode all bboxes in a batch. More...

List< NormalizedBBox >	Decode (List< NormalizedBBox > rgPriorBbox, List< List< float > > rgrgfPriorVariance, PriorBoxParameter.CodeType code_type, bool bEncodeVarianceInTarget, bool bClip, List< NormalizedBBox > rgBbox)
	Decode a set of bounding box. More...

NormalizedBBox	Decode (NormalizedBBox prior_bbox, List< float > rgfPriorVariance, PriorBoxParameter.CodeType code_type, bool bEncodeVarianceInTarget, bool bClip, NormalizedBBox bbox)
	Decode a bounding box. More...

NormalizedBBox	Encode (NormalizedBBox prior_bbox, List< float > rgfPriorVariance, PriorBoxParameter.CodeType code_type, bool bEncodeVarianceInTarget, NormalizedBBox bbox)
	Encode a bounding box. More...

bool	MeetEmitConstraint (NormalizedBBox src_bbox, NormalizedBBox bbox, EmitConstraint emit_constraint)
	Check if a bbox meets the emit constraint w.r.t the src_bbox. More...

float	Coverage (NormalizedBBox bbox1, NormalizedBBox bbox2)
	Compute the coverage of bbox1 by bbox2. More...

NormalizedBBox	Locate (NormalizedBBox srcBbox, NormalizedBBox bbox)
	Locate bbox in the coordinate system of the source Bbox. More...

float	JaccardOverlap (NormalizedBBox bbox1, NormalizedBBox bbox2, bool bNormalized=true)
	Calculates the Jaccard overlap between two bounding boxes. More...

NormalizedBBox	Output (NormalizedBBox bbox, SizeF szImg, ResizeParameter p)
	Output the predicted bbox on the actual image. More...

bool	Project (NormalizedBBox src, NormalizedBBox bbox, out NormalizedBBox proj_bbox)
	Project one bbox onto another. More...

void	Extrapolate (ResizeParameter param, int nHeight, int nWidth, NormalizedBBox crop_bbox, NormalizedBBox bbox)
	Extrapolate the transformed bbox if height_scale and width_scale are explicitly provied, and the FIT_SMALL_SIZE resize mode is specified. More...

NormalizedBBox	Intersect (NormalizedBBox bbox1, NormalizedBBox bbox2)
	Create the intersection of two bounding boxes. More...

NormalizedBBox	Clip (NormalizedBBox bbox, float fHeight=1.0f, float fWidth=1.0f)
	Clip the BBox to a set range. More...

NormalizedBBox	Scale (NormalizedBBox bbox, int nHeight, int nWidth)
	Scale the BBox to a set range. More...

float	Size (NormalizedBBox bbox, bool bNormalized=true)
	Calculate the size of a BBox. More...

void	FindMatches (List< LabelBBox > rgAllLocPreds, DictionaryMap< List< NormalizedBBox > > rgAllGtBboxes, List< NormalizedBBox > rgPriorBboxes, List< List< float > > rgrgPriorVariances, MultiBoxLossParameter p, out List< DictionaryMap< List< float > > > rgAllMatchOverlaps, out List< DictionaryMap< List< int > > > rgAllMatchIndices)
	Find matches between prediction bboxes and ground truth bboxes. More...

int	CountNumMatches (List< DictionaryMap< List< int > > > rgAllMatchIndices, int nNum)
	Counts the number of matches in the list of maps. More...

bool	IsEligibleMining (MultiBoxLossParameter.MiningType miningType, int nMatchIdx, float fMatchOverlap, float fNegOverlap)
	Returns whether or not mining is eligible given the mining type and match index. More...

int	MineHardExamples (Blob< T > blobConf, List< LabelBBox > rgAllLocPreds, DictionaryMap< List< NormalizedBBox > > rgAllGtBBoxes, List< NormalizedBBox > rgPriorBboxes, List< List< float > > rgrgPriorVariances, List< DictionaryMap< List< float > > > rgAllMatchOverlaps, MultiBoxLossParameter p, List< DictionaryMap< List< int > > > rgAllMatchIndices, List< List< int > > rgAllNegIndices, out int nNumNegs)
	Mine the hard examples from the batch. More...

void	EncodeLocPrediction (List< LabelBBox > rgAllLocPreds, DictionaryMap< List< NormalizedBBox > > rgAllGtBboxes, List< DictionaryMap< List< int > > > rgAllMatchIndices, List< NormalizedBBox > rgPriorBboxes, List< List< float > > rgrgPriorVariances, MultiBoxLossParameter p, Blob< T > blobLocPred, Blob< T > blobLocGt)
	Encode the localization prediction and ground truth for each matched prior. More...

void	EncodeConfPrediction (float[] rgfConfData, int nNum, int nNumPriors, MultiBoxLossParameter p, List< DictionaryMap< List< int > > > rgAllMatchIndices, List< List< int > > rgAllNegIndices, DictionaryMap< List< NormalizedBBox > > rgAllGtBBoxes, Blob< T > blobConfPred, Blob< T > blobConfGt)
	Encode the confidence predictions and ground truth for each matched prior. More...

List< List< float > >	ComputeLocLoss (Blob< T > blobLocPred, Blob< T > blobLocGt, List< DictionaryMap< List< int > > > rgAllMatchIndices, int nNum, int nNumPriors, MultiBoxLossParameter.LocLossType lossType)
	Compute the localization loss per matched prior. More...

Detailed Description

The BBox class processes the NormalizedBBox data used with SSD.

See also: SSD: Single Shot MultiBox Detector by Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg, 2016.; GitHub: SSD: Single Shot MultiBox Detector, by weiliu89/caffe, 2016

Definition at line 21 of file BBoxUtility.cs.

Constructor & Destructor Documentation

◆ BBoxUtility()

MyCaffe.common.BBoxUtility< T >.BBoxUtility	(	CudaDnn< T >	cuda,
		Log	log
	)

The constructor.

Parameters

cuda	Specifies the connection to Cuda and cuDNN.
log	Specifies the output log.

Definition at line 33 of file BBoxUtility.cs.

Member Function Documentation

◆ ApplyNMS() [1/2]

List< int > MyCaffe.common.BBoxUtility< T >.ApplyNMS	(	List< NormalizedBBox >	rgBBoxes,
		List< float >	rgScores,
		float	fThreshold,
		int	nTopK
	)

Do non maximum supression given bboxes and scores.

Parameters

rgBBoxes	Specifies a set of bounding boxes.
rgScores	Specifies a seto of corresponding confidences.
fThreshold	Specifies the threshold used in non maximum suppression.
nTopK	Specifies the top k picked indices or -1 for all.

Returns: The indices of the bboxes after nms are returned.

Definition at line 312 of file BBoxUtility.cs.

◆ ApplyNMS() [2/2]

List< int > MyCaffe.common.BBoxUtility< T >.ApplyNMS	(	List< NormalizedBBox >	rgBBoxes,
		List< float >	rgScores,
		float	fThreshold,
		int	nTopK,
		bool	bReuseOverlaps,
		out Dictionary< int, Dictionary< int, float > >	rgOverlaps
	)

Do non maximum supression given bboxes and scores.

Parameters

rgBBoxes	Specifies a set of bounding boxes.
rgScores	Specifies a seto of corresponding confidences.
fThreshold	Specifies the threshold used in non maximum suppression.
nTopK	Specifies the top k picked indices or -1 for all.
bReuseOverlaps	Specifies whether or not to use and update overlaps (true) or alwasy compute the overlap (false).
rgOverlaps	Returns the overlaps between pairs of bboxes if bReuseOverlaps is true.

Returns: The indices of the bboxes after nms are returned.

Definition at line 328 of file BBoxUtility.cs.

◆ ApplyNMSFast()

void MyCaffe.common.BBoxUtility< T >.ApplyNMSFast	(	List< NormalizedBBox >	rgBBoxes,
		List< float >	rgScores,
		float	fScoreThreshold,
		float	fNmsThreshold,
		float	fEta,
		int	nTopK,
		out List< int >	rgIndices
	)

Do a fast non maximum supression given bboxes and scores.

Parameters

rgBBoxes	Specifies a set of bounding boxes.
rgScores	Specifies a seto of corresponding confidences.
fScoreThreshold	Specifies the score threshold used in non maximum suppression.
fNmsThreshold	Specifies the nms threshold used in non maximum suppression.
fEta	Specifies the eta value.
nTopK	Specifies the top k picked indices or -1 for all.
rgIndices	Returns the kept indices of bboxes after nms.

Definition at line 263 of file BBoxUtility.cs.

◆ Clip()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Clip	(	NormalizedBBox	bbox,
		float	fHeight = `1.0f`,
		float	fWidth = `1.0f`
	)

Clip the BBox to a set range.

Parameters

bbox	Specifies the input bounding box.
fHeight	Specifies the clipping height.
fWidth	Specifies the clipping width.

Returns: A new, clipped NormalizedBBox is returned.

Definition at line 1723 of file BBoxUtility.cs.

◆ ComputeAP()

float MyCaffe.common.BBoxUtility< T >.ComputeAP	(	List< Tuple< float, int > >	rgTp,
		int	nNumPos,
		List< Tuple< float, int > >	rgFp,
		ApVersion	apVersion,
		out List< float >	rgPrec,
		out List< float >	rgRec
	)

Compute the average precision given true positive and false positive vectors.

Parameters

rgTp	Specifies a list of scores and true positive.
nNumPos	Specifies the number of true positives.
rgFp	Specifies a list of scores and false positive.
apVersion	Specifies the different ways of computing the Average Precisions.

Tag: Average Precision

Versions: 11point: The 11-point interpolated average precision, used in VOC2007. MaxIntegral: maximally interpolated AP. Used in VOC2012/ILSVRC. Integral: the natrual integral of the precision-recall curve.

Parameters

rgPrec	Returns the computed precisions.
rgRec	Returns the computed recalls.

Returns: The Average Precision value is returned.

Definition at line 69 of file BBoxUtility.cs.

◆ ComputeConfLoss() [1/3]

List< List< float > > MyCaffe.common.BBoxUtility< T >.ComputeConfLoss	(	float[]	rgConfData,
		int	nNum,
		int	nNumPredsPerClass,
		int	nNumClasses,
		int	nBackgroundLabelId,
		MultiBoxLossParameter.ConfLossType	loss_type
	)

Compute the confidence loss for each prior from rgConfData.

Parameters

rgConfData	Specifies the nNum x nNumPredsPerClass * nNumClasses blob of confidence data.
nNum	Specifies the number of images.
nNumPredsPerClass	Specifies the number of predictions per class.
nNumClasses	Specifies the number of classes.
nBackgroundLabelId	Specifies the background label.
loss_type	Specifies the loss type used to compute the confidence.

Returns: The confidence loss values are returned with confidence loss per location for each image.

Definition at line 551 of file BBoxUtility.cs.

◆ ComputeConfLoss() [2/3]

List< List< float > > MyCaffe.common.BBoxUtility< T >.ComputeConfLoss	(	float[]	rgConfData,
		int	nNum,
		int	nNumPredsPerClass,
		int	nNumClasses,
		int	nBackgroundLabelId,
		MultiBoxLossParameter.ConfLossType	loss_type,
		List< Dictionary< int, List< int > > >	rgAllMatchIndices,
		Dictionary< int, List< NormalizedBBox > >	rgAllGtBoxes
	)

Compute the confidence loss for each prior from rgConfData.

Parameters

rgConfData	Specifies the nNum x nNumPredsPerClass * nNumClasses blob of confidence data.
nNum	Specifies the number of images.
nNumPredsPerClass	Specifies the number of predictions per class.
nNumClasses	Specifies the number of classes.
nBackgroundLabelId	Specifies the background label.
loss_type	Specifies the loss type used to compute the confidence.
rgAllMatchIndices	Specifies all match indices storing a mapping between predictions and ground truth.
rgAllGtBoxes	Specifies all ground truth bboxes from the batch.

Returns: The confidence loss values are returned with confidence loss per location for each image.

Definition at line 657 of file BBoxUtility.cs.

◆ ComputeConfLoss() [3/3]

List< List< float > > MyCaffe.common.BBoxUtility< T >.ComputeConfLoss	(	float[]	rgConfData,
		int	nNum,
		int	nNumPredsPerClass,
		int	nNumClasses,
		int	nBackgroundLabelId,
		MultiBoxLossParameter.ConfLossType	loss_type,
		List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		DictionaryMap< List< NormalizedBBox > >	rgAllGtBoxes
	)

Compute the confidence loss for each prior from rgConfData.

Parameters

rgConfData	Specifies the nNum x nNumPredsPerClass * nNumClasses blob of confidence data.
nNum	Specifies the number of images.
nNumPredsPerClass	Specifies the number of predictions per class.
nNumClasses	Specifies the number of classes.
nBackgroundLabelId	Specifies the background label.
loss_type	Specifies the loss type used to compute the confidence.
rgAllMatchIndices	Specifies all match indices storing a mapping between predictions and ground truth.
rgAllGtBoxes	Specifies all ground truth bboxes from the batch.

Returns: The confidence loss values are returned with confidence loss per location for each image.

Definition at line 634 of file BBoxUtility.cs.

◆ ComputeLocLoss()

List< List< float > > MyCaffe.common.BBoxUtility< T >.ComputeLocLoss	(	Blob< T >	blobLocPred,
		Blob< T >	blobLocGt,
		List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		int	nNum,
		int	nNumPriors,
		MultiBoxLossParameter.LocLossType	lossType
	)

Compute the localization loss per matched prior.

Parameters

blobLocPred	Specifies the location prediction results.
blobLocGt	Specifies the encoded location ground truth.
rgAllMatchIndices	Specifies the mapping between predictions and the ground truth.
nNum	Specifies the number of images in the batch.
nNumPriors	Specifies the total number of priors.
lossType	Specifies the type of localization loss, Smooth_L1 or L2.

Returns: Returns the localization loss for all priors in the batch.

Definition at line 2429 of file BBoxUtility.cs.

◆ CountNumMatches()

int MyCaffe.common.BBoxUtility< T >.CountNumMatches	(	List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		int	nNum
	)

Counts the number of matches in the list of maps.

Parameters

rgAllMatchIndices	Specifies the list of match indices.
nNum	Specifies the number of items.

Returns: The total matches found in the number of items is returned.

Definition at line 1902 of file BBoxUtility.cs.

◆ Coverage()

float MyCaffe.common.BBoxUtility< T >.Coverage	(	NormalizedBBox	bbox1,
		NormalizedBBox	bbox2
	)

Compute the coverage of bbox1 by bbox2.

Parameters

bbox1	Specifies the first BBox.
bbox2	Specifies the second BBox.

Returns: The coverage of bbox1 by bbox2 is returned.

Definition at line 1489 of file BBoxUtility.cs.

◆ CumSum()

List< int > MyCaffe.common.BBoxUtility< T >.CumSum ( List< Tuple< float, int > > rgPairs )

Calculate the cumulative sum of a set of pairs.

Parameters

rgPairs

Returns

Definition at line 187 of file BBoxUtility.cs.

◆ Decode() [1/2]

List< NormalizedBBox > MyCaffe.common.BBoxUtility< T >.Decode	(	List< NormalizedBBox >	rgPriorBbox,
		List< List< float > >	rgrgfPriorVariance,
		PriorBoxParameter.CodeType	code_type,
		bool	bEncodeVarianceInTarget,
		bool	bClip,
		List< NormalizedBBox >	rgBbox
	)

Decode a set of bounding box.

Parameters

rgPriorBbox	Specifies an list of prior bounding boxs.
rgrgfPriorVariance	Specifies the list of prior variance (must have 4 elements each of which are > 0).
code_type	Specifies the code type.
bEncodeVarianceInTarget	Specifies whether or not to encode the variance in the target.
bClip	Specifies to enable/disable the clipping.
rgBbox	Specifies a list of bounding boxs.

Returns: A list of decoded bounding box is returned.

Definition at line 1185 of file BBoxUtility.cs.

◆ Decode() [2/2]

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Decode	(	NormalizedBBox	prior_bbox,
		List< float >	rgfPriorVariance,
		PriorBoxParameter.CodeType	code_type,
		bool	bEncodeVarianceInTarget,
		bool	bClip,
		NormalizedBBox	bbox
	)

Decode a bounding box.

Parameters

prior_bbox	Specifies the prior bounding box.
rgfPriorVariance	Specifies the prior variance (must have 4 elements each of which are > 0).
code_type	Specifies the code type.
bEncodeVarianceInTarget	Specifies whether or not to encode the variance in the target.
bClip	Specifies whether or not to enable clip or not.
bbox	Specifies the bounding box.

Returns: The decoded bounding box is returned.

Definition at line 1215 of file BBoxUtility.cs.

◆ DecodeAll()

List< LabelBBox > MyCaffe.common.BBoxUtility< T >.DecodeAll	(	List< LabelBBox >	rgAllLocPreds,
		List< NormalizedBBox >	rgPriorBboxes,
		List< List< float > >	rgrgfPrioVariances,
		int	nNum,
		bool	bShareLocation,
		int	nNumLocClasses,
		int	nBackgroundLabelId,
		PriorBoxParameter.CodeType	codeType,
		bool	bVarianceEncodedInTarget,
		bool	bClip
	)

Decode all bboxes in a batch.

Parameters

rgAllLocPreds	Specifies the batch of local predictions.
rgPriorBboxes	Specifies the set of prior bboxes.
rgrgfPrioVariances	Specifies the prior variances.
nNum	Specifies the number of items in the batch.
bShareLocation	Specifies whether or not to share locations.
nNumLocClasses	Specifies the number of local classes.
nBackgroundLabelId	Specifies the background label.
codeType	Specifies the coding type.
bVarianceEncodedInTarget	Specifies whether or not the variance is encoded in the target or not.
bClip	Specifies whether or not to clip.

Returns: The decoded Bboxes are returned.

Definition at line 1142 of file BBoxUtility.cs.

◆ Dispose()

void MyCaffe.common.BBoxUtility< T >.Dispose ( )

Clean up all resources.

Definition at line 43 of file BBoxUtility.cs.

◆ Encode()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Encode	(	NormalizedBBox	prior_bbox,
		List< float >	rgfPriorVariance,
		PriorBoxParameter.CodeType	code_type,
		bool	bEncodeVarianceInTarget,
		NormalizedBBox	bbox
	)

Encode a bounding box.

Parameters

prior_bbox	Specifies the prior bounding box.
rgfPriorVariance	Specifies the prior variance (must have 4 elements each of which are > 0).
code_type	Specifies the code type.
bEncodeVarianceInTarget	Specifies whether or not to encode the variance in the target.
bbox	Specifies the bounding box.

Returns: The encoded bounding box is returned.

Definition at line 1337 of file BBoxUtility.cs.

◆ EncodeConfPrediction()

void MyCaffe.common.BBoxUtility< T >.EncodeConfPrediction	(	float[]	rgfConfData,
		int	nNum,
		int	nNumPriors,
		MultiBoxLossParameter	p,
		List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		List< List< int > >	rgAllNegIndices,
		DictionaryMap< List< NormalizedBBox > >	rgAllGtBBoxes,
		Blob< T >	blobConfPred,
		Blob< T >	blobConfGt
	)

Encode the confidence predictions and ground truth for each matched prior.

Parameters

rgfConfData	Specifies the num x num_priors * num_classes blob.
nNum	Specifies the number of images.
nNumPriors	Specifies the number of priors (predictions) per image.
p	Specifies the parameters for the MultiBoxLossLayer.
rgAllMatchIndices	Specifies the mapping between predictions and the ground truth.
rgAllNegIndices	Specifies the indices for negative samples.
rgAllGtBBoxes	Specifies the ground truth bboxes for the batch.
blobConfPred	Specifies the confidence prediction results.
blobConfGt	Specifies the confidence ground truth.

Definition at line 2297 of file BBoxUtility.cs.

◆ EncodeLocPrediction()

void MyCaffe.common.BBoxUtility< T >.EncodeLocPrediction	(	List< LabelBBox >	rgAllLocPreds,
		DictionaryMap< List< NormalizedBBox > >	rgAllGtBboxes,
		List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		List< NormalizedBBox >	rgPriorBboxes,
		List< List< float > >	rgrgPriorVariances,
		MultiBoxLossParameter	p,
		Blob< T >	blobLocPred,
		Blob< T >	blobLocGt
	)

Encode the localization prediction and ground truth for each matched prior.

Parameters

rgAllLocPreds	Specifies the location prediction, where each item contains the location prediction for an image.
rgAllGtBboxes	Specifies the ground truth bboxes for the batch.
rgAllMatchIndices	Specifies the mapping between predictions and the ground truth.
rgPriorBboxes	Specifies the prior bboxes in the format of NormalizedBBox.
rgrgPriorVariances	Specifies the variances needed by the prior bboxes.
p	Specifies the parameters for the MultiBoxLossLayer.
blobLocPred	Specifies the location prediction results.
blobLocGt	Specifies the encoded location ground truth.

Definition at line 2195 of file BBoxUtility.cs.

◆ Extrapolate()

void MyCaffe.common.BBoxUtility< T >.Extrapolate	(	ResizeParameter	param,
		int	nHeight,
		int	nWidth,
		NormalizedBBox	crop_bbox,
		NormalizedBBox	bbox
	)

Extrapolate the transformed bbox if height_scale and width_scale are explicitly provied, and the FIT_SMALL_SIZE resize mode is specified.

Parameters

param	Specifies the resize parameter.
nHeight	Specifies the height.
nWidth	Specifies the width.
crop_bbox	Specifies the crop Bbox.
bbox	Specifies the Bbox to be updated.

Definition at line 1668 of file BBoxUtility.cs.

◆ FindMatches()

void MyCaffe.common.BBoxUtility< T >.FindMatches	(	List< LabelBBox >	rgAllLocPreds,
		DictionaryMap< List< NormalizedBBox > >	rgAllGtBboxes,
		List< NormalizedBBox >	rgPriorBboxes,
		List< List< float > >	rgrgPriorVariances,
		MultiBoxLossParameter	p,
		out List< DictionaryMap< List< float > > >	rgAllMatchOverlaps,
		out List< DictionaryMap< List< int > > >	rgAllMatchIndices
	)

Find matches between prediction bboxes and ground truth bboxes.

Parameters

rgAllLocPreds	Specifies the location prediction, where each item contains location prediction for an image.
rgAllGtBboxes	Specifies the ground truth bboxes for the batch.
rgPriorBboxes	Specifies all prior bboxes in the format of NormalizedBBox.
rgrgPriorVariances	Specifies all the variances needed by prior bboxes.
p	Specifies the parameter for the MultiBoxLossLayer.
rgAllMatchOverlaps	Returns the jaccard overlaps between predictions and ground truth.
rgAllMatchIndices	Returns the mapping between predictions and ground truth.

Definition at line 1786 of file BBoxUtility.cs.

◆ GetConfidenceScores()

List< Dictionary< int, List< float > > > MyCaffe.common.BBoxUtility< T >.GetConfidenceScores	(	float[]	rgConfData,
		int	nNum,
		int	nNumPredsPerClass,
		int	nNumClasses
	)

Calculate the confidence predictions from rgConfData.

Parameters

rgConfData	Specifies the nNum x nNumPredsPerClass * nNumClasses blob of confidence data.
nNum	Specifies the number of images.
nNumPredsPerClass	Specifies the number of predictions per class.
nNumClasses	Specifies the number of classes.

Returns: The confidence scores are returned as the confidence predictions which contains a confidence prediction for an image.

Definition at line 760 of file BBoxUtility.cs.

◆ GetDetectionResults()

Dictionary< int, Dictionary< int, List< NormalizedBBox > > > MyCaffe.common.BBoxUtility< T >.GetDetectionResults	(	float[]	rgData,
		int	nNumDet,
		int	nBackgroundLabelId
	)

Get detection results from rgData.

Parameters

rgData	Specifies a 1 x 1 x nNumDet x 7 blob data.
nNumDet	Specifies the number of detections.
nBackgroundLabelId	Specifies the label for the background class which is used to do a sanity check so that no detection contains it.

Returns: The detection results are returned for each class from each image.

Definition at line 435 of file BBoxUtility.cs.

◆ GetGroundTruth()

DictionaryMap< List< NormalizedBBox > > MyCaffe.common.BBoxUtility< T >.GetGroundTruth	(	float[]	rgGtData,
		int	nNumGt,
		int	nBackgroundLabelId,
		bool	bUseDifficultGt
	)

Create a set of ground truth bounding boxes from the rgGtData.

Parameters

rgGtData	Specifies the 1 x 1 x nNumGt x 7 blob with ground truth initialization data.
nNumGt	Specifies the number of ground truths.
nBackgroundLabelId	Specifies the background label.
bUseDifficultGt	Specifies whether or not to use the difficult ground truth.

Returns: A dictionary containing the ground truth's (one per image) is returned with the label of each bbox stored in the NormalizedBBox.

Definition at line 844 of file BBoxUtility.cs.

◆ GetGroundTruthEx()

Dictionary< int, LabelBBox > MyCaffe.common.BBoxUtility< T >.GetGroundTruthEx	(	float[]	rgGtData,
		int	nNumGt,
		int	nBackgroundLabelId,
		bool	bUseDifficultGt
	)

Create a set of ground truth bounding boxes from the rgGtData.

Parameters

rgGtData	Specifies the 1 x 1 x nNumGt x 7 blob with ground truth initialization data.
nNumGt	Specifies the number of ground truths.
nBackgroundLabelId	Specifies the background label.
bUseDifficultGt	Specifies whether or not to use the difficult ground truth.

Returns: A dictionary containing the ground truths per label is returned.

Definition at line 888 of file BBoxUtility.cs.

◆ GetLocPredictions()

List< LabelBBox > MyCaffe.common.BBoxUtility< T >.GetLocPredictions	(	float[]	rgLocData,
		int	nNum,
		int	nNumPredsPerClass,
		int	nNumLocClasses,
		bool	bShareLocation
	)

Create a set of local predictions from the rgLocData.

Parameters

rgLocData	Specifies the nNum x nNumPredsPerClass * nNumLocClasses * 4 blbo with prediction initialization data.
nNum	Specifies the number of images.
nNumPredsPerClass	Specifies the number of predictions per class.
nNumLocClasses	Specifies the number of local classes It is 1 if bShareLocation is true; and it is equal to the number of classes needed to predict otherwise.
bShareLocation	Specifies whether or not to share the location. If true, all classes share the same location prediction.

Returns: A list of created location predictions is returned as a list of LabelBBox items where each item contains a location prediction for the image.

Definition at line 802 of file BBoxUtility.cs.

◆ GetPrior()

List< NormalizedBBox > MyCaffe.common.BBoxUtility< T >.GetPrior	(	float[]	rgPriorData,
		int	nNumPriors,
		out List< List< float > >	rgPriorVariances
	)

Get the prior boundary boxes from the rgPriorData.

Parameters

rgPriorData	Specifies the prior data as a 1 x 2 x nNumPriors x 4 x 1 blob.
nNumPriors	Specifies the number of priors.
rgPriorVariances	Specifies the prior variances need by prior bboxes.

Returns: The prior bbox list is returned.

Definition at line 477 of file BBoxUtility.cs.

◆ Intersect()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Intersect	(	NormalizedBBox	bbox1,
		NormalizedBBox	bbox2
	)

Create the intersection of two bounding boxes.

Parameters

bbox1	Specifies the first bounding box.
bbox2	Specifies the second bounding box.

Returns: The intersection of the two bounding boxes is returned.

Definition at line 1703 of file BBoxUtility.cs.

◆ IsCrossBoundary()

bool MyCaffe.common.BBoxUtility< T >.IsCrossBoundary ( NormalizedBBox bbox )

Returns whether or not the bbox is overlaps outside the range [0,1]

Parameters

bbox	Specifies the bounding box to test.

Returns: If the bbox overlaps, true is returned, otherwise false is returned.

Definition at line 1111 of file BBoxUtility.cs.

◆ IsEligibleMining()

bool MyCaffe.common.BBoxUtility< T >.IsEligibleMining	(	MultiBoxLossParameter.MiningType	miningType,
		int	nMatchIdx,
		float	fMatchOverlap,
		float	fNegOverlap
	)

Returns whether or not mining is eligible given the mining type and match index.

Parameters

miningType	Specifies the mining type.
nMatchIdx	Specifies the matching index.
fMatchOverlap	Specifies the matching overlap.
fNegOverlap	Specifies the negative overlap.

Returns: If mining is allowed, true is returned, otherwise false is returned.

Definition at line 1933 of file BBoxUtility.cs.

◆ JaccardOverlap()

float MyCaffe.common.BBoxUtility< T >.JaccardOverlap	(	NormalizedBBox	bbox1,
		NormalizedBBox	bbox2,
		bool	bNormalized = `true`
	)

Calculates the Jaccard overlap between two bounding boxes.

Parameters

bbox1	Specifies the first bounding box.
bbox2	Specifies the second bounding box.
bNormalized	Specifies whether or not the bboxes are normalized or not.

Returns: The Jaccard overlap is returned.

Definition at line 1527 of file BBoxUtility.cs.

◆ Locate()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Locate	(	NormalizedBBox	srcBbox,
		NormalizedBBox	bbox
	)

Locate bbox in the coordinate system of the source Bbox.

Parameters

srcBbox	Specifies the source Bbox.
bbox	Specifies the bbox to locate.

Returns: The bbox located within the source Bbox is returned.

Definition at line 1509 of file BBoxUtility.cs.

◆ Match()

void MyCaffe.common.BBoxUtility< T >.Match	(	List< NormalizedBBox >	rgGtBboxes,
		List< NormalizedBBox >	rgPredBboxes,
		int	nLabel,
		MultiBoxLossParameter.MatchType	match_type,
		float	fOverlapThreshold,
		bool	bIgnoreCrossBoundaryBbox,
		out List< int >	rgMatchIndices,
		out List< float >	rgMatchOverlaps
	)

Find matches between a list of two bounding boxes.

Parameters

rgGtBboxes	Specifies a list of ground truth bounding boxes.
rgPredBboxes	Specifies a list of predicted bounding boxes.
nLabel	Specifies the label.
match_type	Specifies the matching type.
fOverlapThreshold	Specifies the overlap.
bIgnoreCrossBoundaryBbox	Specifies whether or not to ignore corss boundary bounding boxes.
rgMatchIndices	Specifies the list where the indexes of matches are placed.
rgMatchOverlaps	Specifies the list where the overlaps of matches are placed.

Definition at line 935 of file BBoxUtility.cs.

◆ MeetEmitConstraint()

bool MyCaffe.common.BBoxUtility< T >.MeetEmitConstraint	(	NormalizedBBox	src_bbox,
		NormalizedBBox	bbox,
		EmitConstraint	emit_constraint
	)

Check if a bbox meets the emit constraint w.r.t the src_bbox.

Parameters

src_bbox	Specifies the source Bbox.
bbox	Specifies the Bbox to check.
emit_constraint	Specifies the emit constraint.

Returns: If the emit constraint is met, true is returned, otherwise false is returned.

Definition at line 1455 of file BBoxUtility.cs.

◆ MineHardExamples()

int MyCaffe.common.BBoxUtility< T >.MineHardExamples	(	Blob< T >	blobConf,
		List< LabelBBox >	rgAllLocPreds,
		DictionaryMap< List< NormalizedBBox > >	rgAllGtBBoxes,
		List< NormalizedBBox >	rgPriorBboxes,
		List< List< float > >	rgrgPriorVariances,
		List< DictionaryMap< List< float > > >	rgAllMatchOverlaps,
		MultiBoxLossParameter	p,
		List< DictionaryMap< List< int > > >	rgAllMatchIndices,
		List< List< int > >	rgAllNegIndices,
		out int	nNumNegs
	)

Mine the hard examples from the batch.

Parameters

blobConf	Specifies the confidence prediction.
rgAllLocPreds	Specifies the location prediction, where each item contains the location prediction for an image.
rgAllGtBBoxes	Specifies the ground truth bboxes for the batch.
rgPriorBboxes	Specifies the prior bboxes in the format of NormalizedBBox.
rgrgPriorVariances	Specifies the variances needed by the prior bboxes.
rgAllMatchOverlaps	Specifies the jaccard overlap between predictions and the ground truth.
p	Specifies the parameters for the MultiBoxLossLayer.
rgAllMatchIndices	Specifies the mapping between predictions and the ground truth.
rgAllNegIndices	Specifies the indices for negative samples.
nNumNegs	Specifies the numberof negative indices.

Returns: The number of matches is returned.

Definition at line 1966 of file BBoxUtility.cs.

◆ Output()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Output	(	NormalizedBBox	bbox,
		SizeF	szImg,
		ResizeParameter	p
	)

Output the predicted bbox on the actual image.

Parameters

bbox	Specifies the bbox.
szImg	Specifies the image size.
p	Specifies the resize parameter.

Returns: The predicted bbox is returned.

Definition at line 1557 of file BBoxUtility.cs.

◆ Project()

bool MyCaffe.common.BBoxUtility< T >.Project	(	NormalizedBBox	src,
		NormalizedBBox	bbox,
		out NormalizedBBox	proj_bbox
	)

Project one bbox onto another.

Parameters

src	Specifies the source bbox.
bbox	Specifies the second bbox.
proj_bbox	Returns the projected bbox here.

Returns: The new project bbox is returned if a projection was made, otherwise the original bbox is returned.

Definition at line 1635 of file BBoxUtility.cs.

◆ Scale()

NormalizedBBox MyCaffe.common.BBoxUtility< T >.Scale	(	NormalizedBBox	bbox,
		int	nHeight,
		int	nWidth
	)

Scale the BBox to a set range.

Parameters

bbox	Specifies the input bounding box.
nHeight	Specifies the scaling height.
nWidth	Specifies the scaling width.

Returns: A new, scaled NormalizedBBox is returned.

Definition at line 1741 of file BBoxUtility.cs.

◆ Size()

float MyCaffe.common.BBoxUtility< T >.Size	(	NormalizedBBox	bbox,
		bool	bNormalized = `true`
	)

Calculate the size of a BBox.

Parameters

bbox	Specifies the input bounding box.
bNormalized	Specifies whether or not the bounding box falls within the range [0,1].

Returns: The size of the bounding box is returned.

Definition at line 1759 of file BBoxUtility.cs.

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

C:/Data/Data/SS_Projects/Intelligence/GitHub/MyCaffe/MyCaffe/common/BBoxUtility.cs

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ BBoxUtility()

Member Function Documentation

◆ ApplyNMS() [1/2]

◆ ApplyNMS() [2/2]

◆ ApplyNMSFast()

◆ Clip()

◆ ComputeAP()

◆ ComputeConfLoss() [1/3]

◆ ComputeConfLoss() [2/3]

◆ ComputeConfLoss() [3/3]

◆ ComputeLocLoss()

◆ CountNumMatches()

◆ Coverage()

◆ CumSum()

◆ Decode() [1/2]

◆ Decode() [2/2]

◆ DecodeAll()

◆ Dispose()

◆ Encode()

◆ EncodeConfPrediction()

◆ EncodeLocPrediction()

◆ Extrapolate()

◆ FindMatches()

◆ GetConfidenceScores()

◆ GetDetectionResults()

◆ GetGroundTruth()

◆ GetGroundTruthEx()

◆ GetLocPredictions()

◆ GetPrior()

◆ Intersect()

◆ IsCrossBoundary()

◆ IsEligibleMining()

◆ JaccardOverlap()

◆ Locate()

◆ Match()

◆ MeetEmitConstraint()

◆ MineHardExamples()

◆ Output()

◆ Project()

◆ Scale()

◆ Size()