Package ai.djl.training.loss

Class YOLOv3Loss

java.lang.Object

ai.djl.training.evaluator.Evaluator

ai.djl.training.loss.Loss

ai.djl.training.loss.YOLOv3Loss

public final class YOLOv3Loss
extends Loss

YOLOv3Loss is an implementation of Loss. It is used to compute the loss while training a YOLOv3 model for object detection. It involves computing the targets given the generated anchors, labels and predictions, and then computing the sum of class predictions and bounding box predictions.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	YOLOv3Loss.Builder	The Builder to construct a YOLOv3Loss object.

Field Summary

Fields inherited from class ai.djl.training.evaluator.Evaluator

totalInstances

Method Summary

Modifier and Type	Method	Description
NDArray	bceLoss(NDArray prediction, NDArray target)	Calculates the BCELoss between prediction and target.
static YOLOv3Loss.Builder	builder()	Creates a new builder to build a YOLOv3Loss.
NDArray	calculateIOU(NDArray predx, NDArray predy, NDArray groundTruth, int componentIndex)	Calculates the IOU between priori Anchors and groundTruth.
NDArray	clipByTensor(NDArray tList, float tMin, float tMax)	Make the value of given NDArray between tMin and tMax.
NDArray	evaluate(NDList labels, NDList predictions)	Calculates the evaluation between the labels and the predictions.
NDArray	evaluateOneOutput(int componentIndex, NDArray input, NDArray labels)	Computes the Loss for one outputLayer.
static float[]	getPresetAnchors()	Gets the preset anchors of YoloV3.
NDList	getTarget(NDArray labels, int inH, int inW)	Gets target NDArray for a given evaluator.
NDArray	mseLoss(NDArray prediction, NDArray target)	Calculates the MSELoss between prediction and target.

Methods inherited from class ai.djl.training.loss.Loss

addAccumulator, elasticNetWeightedDecay, elasticNetWeightedDecay, elasticNetWeightedDecay, elasticNetWeightedDecay, getAccumulator, hingeLoss, hingeLoss, hingeLoss, l1Loss, l1Loss, l1Loss, l1WeightedDecay, l1WeightedDecay, l1WeightedDecay, l2Loss, l2Loss, l2Loss, l2WeightedDecay, l2WeightedDecay, l2WeightedDecay, maskedSoftmaxCrossEntropyLoss, maskedSoftmaxCrossEntropyLoss, maskedSoftmaxCrossEntropyLoss, quantileL1Loss, quantileL1Loss, resetAccumulator, sigmoidBinaryCrossEntropyLoss, sigmoidBinaryCrossEntropyLoss, sigmoidBinaryCrossEntropyLoss, softmaxCrossEntropyLoss, softmaxCrossEntropyLoss, softmaxCrossEntropyLoss, updateAccumulator

Methods inherited from class ai.djl.training.evaluator.Evaluator

checkLabelShapes, checkLabelShapes, getName

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

getPresetAnchors

public static float[] getPresetAnchors()

Gets the preset anchors of YoloV3.

Returns:

the preset anchors of YoloV3

clipByTensor

public NDArray clipByTensor(NDArray tList,
                            float tMin,
                            float tMax)

Make the value of given NDArray between tMin and tMax.

Parameters:

tList - the given NDArray

tMin - the min value

tMax - the max value

Returns:

a NDArray where values are set between tMin and tMax

mseLoss

public NDArray mseLoss(NDArray prediction,
                       NDArray target)

Calculates the MSELoss between prediction and target.

Parameters:

prediction - the prediction array

target - the target array

Returns:

the MSELoss between prediction and target

bceLoss

public NDArray bceLoss(NDArray prediction,
                       NDArray target)

Calculates the BCELoss between prediction and target.

Parameters:

prediction - the prediction array

target - the target array

Returns:

the BCELoss between prediction and target

evaluate

public NDArray evaluate(NDList labels,
                        NDList predictions)

Calculates the evaluation between the labels and the predictions.

Specified by:

evaluate in class Evaluator

Parameters:

labels - the correct values

predictions - the predicted values

Returns:

the evaluation result

evaluateOneOutput

public NDArray evaluateOneOutput(int componentIndex,
                                 NDArray input,
                                 NDArray labels)

Computes the Loss for one outputLayer.

Parameters:

componentIndex - which outputLayer does current input represent. the shape should be (13*13,26*26,52*52)

input - one prediction layer of YOLOv3

labels - target labels. Must contain (offsetLabels, masks, classlabels)

Returns:

the total loss of a outputLayer

getTarget

public NDList getTarget(NDArray labels,
                        int inH,
                        int inW)

Gets target NDArray for a given evaluator.

Parameters:

labels - the true labels

inH - the height of current layer

inW - the width of current layer

Returns:

an NDList of {boxLossScale and groundTruth}

calculateIOU

public NDArray calculateIOU(NDArray predx,
                            NDArray predy,
                            NDArray groundTruth,
                            int componentIndex)

Calculates the IOU between priori Anchors and groundTruth.

Parameters:

predx - the tx value of prediction

predy - the ty value of prediction

groundTruth - the groundTruth value of labels

componentIndex - the current component Index

Returns:

an NDArray of IOU

builder

public static YOLOv3Loss.Builder builder()

Creates a new builder to build a YOLOv3Loss.

Returns:

a new builder;