Package ai.djl.modality.cv.translator

Class ObjectDetectionTranslator.ObjectDetectionBuilder<T extends ObjectDetectionTranslator.ObjectDetectionBuilder>

java.lang.Object

ai.djl.modality.cv.translator.BaseImageTranslator.BaseBuilder<T>

ai.djl.modality.cv.translator.BaseImageTranslator.ClassificationBuilder<T>

ai.djl.modality.cv.translator.ObjectDetectionTranslator.ObjectDetectionBuilder<T>

Direct Known Subclasses:

SingleShotDetectionTranslator.Builder, YoloTranslator.Builder, YoloV5Translator.Builder

Enclosing class:

ObjectDetectionTranslator

public abstract static class ObjectDetectionTranslator.ObjectDetectionBuilder<T extends ObjectDetectionTranslator.ObjectDetectionBuilder>
extends BaseImageTranslator.ClassificationBuilder<T>

The base builder for the object detection translator.

Field Summary

Fields

Modifier and Type	Field	Description
protected boolean	applyRatio
protected double	imageHeight
protected double	imageWidth
protected float	threshold

Fields inherited from class ai.djl.modality.cv.translator.BaseImageTranslator.ClassificationBuilder

synsetLoader

Fields inherited from class ai.djl.modality.cv.translator.BaseImageTranslator.BaseBuilder

batchifier, flag, height, pipeline, width

Constructor Summary

Constructors

Constructor	Description
ObjectDetectionBuilder()

Method Summary

Modifier and Type	Method	Description
protected void	configPostProcess(Map<String,?> arguments)
double	getImageHeight()	Get resized image height.
double	getImageWidth()	Get resized image width.
T	optApplyRatio(boolean value)	Determine Whether to divide output object width/height on the inference result.
T	optRescaleSize(double imageWidth, double imageHeight)	Sets the optional rescale size.
T	optThreshold(float threshold)	Sets the threshold for prediction accuracy.

Methods inherited from class ai.djl.modality.cv.translator.BaseImageTranslator.ClassificationBuilder

optSynset, optSynsetArtifactName, optSynsetUrl, validate

Methods inherited from class ai.djl.modality.cv.translator.BaseImageTranslator.BaseBuilder

addTransform, configPreProcess, optBatchifier, optFlag, self, setPipeline

Methods inherited from class java.lang.Object

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

Field Details

threshold

protected float threshold

imageWidth

protected double imageWidth

imageHeight

protected double imageHeight

applyRatio

protected boolean applyRatio

Constructor Details

ObjectDetectionBuilder

public ObjectDetectionBuilder()

Method Details

optThreshold

public T optThreshold(float threshold)

Sets the threshold for prediction accuracy.

Predictions below the threshold will be dropped.

Parameters:

threshold - the threshold for the prediction accuracy

Returns:

this builder

optRescaleSize

public T optRescaleSize(double imageWidth,
 double imageHeight)

Sets the optional rescale size.

Parameters:

imageWidth - the width to rescale images to

imageHeight - the height to rescale images to

Returns:

this builder

optApplyRatio

public T optApplyRatio(boolean value)

Determine Whether to divide output object width/height on the inference result. Default false.

DetectedObject value should always bring a ratio based on the width/height instead of actual width/height. Most of the model will produce ratio as the inference output. This function is aimed to cover those who produce the pixel value. Make this to true to divide the width/height in postprocessing in order to get ratio in detectedObjects.

Parameters:

value - whether to apply ratio

Returns:

this builder

getImageWidth

public double getImageWidth()

Get resized image width.

Returns:

image width

getImageHeight

public double getImageHeight()

Get resized image height.

Returns:

image height

configPostProcess

protected void configPostProcess(Map<String,?> arguments)

Overrides:

configPostProcess in class BaseImageTranslator.ClassificationBuilder<T extends ObjectDetectionTranslator.ObjectDetectionBuilder>