ai.djl.training.loss

Class SingleShotDetectionLoss

java.lang.Object

ai.djl.training.metrics.TrainingMetric

ai.djl.training.loss.Loss

ai.djl.training.loss.SingleShotDetectionLoss

All Implemented Interfaces:

java.lang.Cloneable

public class SingleShotDetectionLoss
extends Loss

SingleShotDetectionLoss is an implementation of Loss. It is used to compute the loss while training a Single Shot Detection (SSD) 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.

Constructor Summary

Constructors

Constructor and Description
SingleShotDetectionLoss(java.lang.String name) Base class for metric with abstract update methods.

Method Summary

Modifier and Type	Method and Description
NDArray	getLoss(NDList labels, NDList predictions) Calculate loss between label and prediction.

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

duplicate, excludeBatchAxis, getValue, hingeLoss, hingeLoss, l1Loss, l1Loss, l2Loss, l2Loss, reset, sigmoidBinaryCrossEntropyLoss, sigmoidBinaryCrossEntropyLoss, softmaxCrossEntropyLoss, softmaxCrossEntropyLoss, update

Methods inherited from class ai.djl.training.metrics.TrainingMetric

checkLabelShapes, checkLabelShapes, getName

Methods inherited from class java.lang.Object

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

Constructor Detail

SingleShotDetectionLoss

public SingleShotDetectionLoss(java.lang.String name)

Base class for metric with abstract update methods.

Parameters:

name - The display name of the Loss

Method Detail

getLoss

public NDArray getLoss(NDList labels,
                       NDList predictions)

Calculate loss between label and prediction.

the default implementation is simply adding all losses together

Specified by:

getLoss in class Loss

Parameters:

labels - target labels. Must contain (offsetLabels, masks, classlabels). This is returned by MultiBoxTarget function

predictions - predicted labels (class prediction, offset prediction)

Returns:

loss value