Class MeanIoU<T extends TNumber>

java.lang.Object

org.tensorflow.framework.metrics.BaseMetric

org.tensorflow.framework.metrics.MeanIoU<T>

Type Parameters:

T - The data type for the metric result

All Implemented Interfaces:

Metric

public class MeanIoU<T extends TNumber>
extends BaseMetric

Computes the mean Intersection-Over-Union metric.

Mean Intersection-Over-Union is a common evaluation metric for semantic image segmentation, which first computes the IOU for each semantic class and then computes the average over classes. IOU is defined as follows: IOU = true_positive / (true_positive + false_positive + false_negative). The predictions are accumulated in a confusion matrix, weighted by sample_weight and the metric is then calculated from it.

If sampleWeight is null, weights default to 1. Use sample_weight of 0 to mask values.

Field Summary

Fields

Modifier and Type	Field	Description
static final String	TOTAL_CONFUSION_MATRIX

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	MeanIoU(long numClasses, long seed, Class<T> type)	Creates a metric MeanIoU, using name as Class.getSimpleName()
protected	MeanIoU(String name, long numClasses, long seed, Class<T> type)	Creates a MeanIoU metric

Method Summary

Modifier and Type	Method	Description
protected void	init(Ops tf)	Initialize the TensorFlow Ops
Op	resetStates(Ops tf)	Resets any state variables to their initial values
<U extends TNumber> Operand<U>	result(Ops tf, Class<U> resultType)	Gets the current result of the metric
List<Op>	updateStateList(Ops tf, Operand<? extends TNumber> labels, Operand<? extends TNumber> predictions, Operand<? extends TNumber> sampleWeights)	Accumulates the confusion matrix statistics.

Methods inherited from class BaseMetric

callOnce, checkIsGraph, getName, getSeed, getTF, getVariableName, isInitialized, setInitialized, setName, setTF, updateState, updateState, updateStateList

Modifier and Type	Method	Description
final <T extends TNumber> Operand<T>	callOnce(Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights, Class<T> type)	Calls update state once, followed by a call to get the result
protected void	checkIsGraph(Ops tf)	Checks if the TensorFlow Ops encapsulates a Graph environment.
String	getName()	The name for this metric.
long	getSeed()	Gets the random number generator seed value
protected Ops	getTF()	Gets the TensorFlow Ops for this metric
protected String	getVariableName(String varName)	Gets a formatted name for a variable, in the form BaseMetric.name + "_" + varName.
boolean	isInitialized()	Checks whether the Metric is initialized or not.
protected void	setInitialized(boolean initialized)	Sets the initialized indicator
void	setName(String name)	Sets the metric name
protected void	setTF(Ops tf)	Sets the TensorFlow Ops for this metric.
final Op	updateState(Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights)	Creates a NoOp Operation with control dependencies to update the metric state
final Op	updateState(Ops tf, Operand<? extends TNumber> labels, Operand<? extends TNumber> predictions, Operand<? extends TNumber> sampleWeights)	Creates a NoOp Operation with control dependencies to update the metric state
List<Op>	updateStateList(Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights)	Creates a List of Operations to update the metric state based on input values.

Methods inherited from class Object

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

Field Details

TOTAL_CONFUSION_MATRIX

public static final String TOTAL_CONFUSION_MATRIX

See Also:

• Constant Field Values

Constructor Details

MeanIoU

protected MeanIoU(long numClasses,
 long seed,
 Class<T> type)

Creates a metric MeanIoU, using name as Class.getSimpleName()

Parameters:

numClasses - The possible number of labels the prediction task can have

seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.

type - the data type for the variables

MeanIoU

protected MeanIoU(String name,
 long numClasses,
 long seed,
 Class<T> type)

Creates a MeanIoU metric

Parameters:

name - the name of the metric, if null then Class.getSimpleName() is used

numClasses - The possible number of labels the prediction task can have

seed - the seed for random number generation. An initializer created with a given seed will always produce the same random tensor for a given shape and data type.

type - the data type for the variables

Method Details

init

protected void init(Ops tf)

Initialize the TensorFlow Ops

Specified by:

init in class BaseMetric

Parameters:

tf - the TensorFlow Ops encapsulating a Graph environment.

resetStates

public Op resetStates(Ops tf)

Resets any state variables to their initial values

Parameters:

tf - the TensorFlow Ops encapsulating a Graph environment.

Returns:

the operation for doing the reset

updateStateList

public List<Op> updateStateList(Ops tf,
 Operand<? extends TNumber> labels,
 Operand<? extends TNumber> predictions,
 Operand<? extends TNumber> sampleWeights)

Accumulates the confusion matrix statistics.

Specified by:

updateStateList in interface Metric

Overrides:

updateStateList in class BaseMetric

Parameters:

tf - the TensorFlow Ops encapsulating a Graph environment.

labels - the labels

predictions - the predictions

sampleWeights - Optional weighting of each example. Defaults to 1, if null. Rank is either 0, or the same rank as labels, and must be broadcastable to labels.

Returns:

the Operands that updates totalConfusionMatrix variable

Throws:

IllegalArgumentException - if the TensorFlow Ops scope does not encapsulate a Graph environment.

IllegalArgumentException - if the weights rank is not 0, and weights rank @{code !=} labels rank, and if the predictions size is not equal to the labels size

result

public <U extends TNumber> Operand<U> result(Ops tf,
 Class<U> resultType)

Gets the current result of the metric

Type Parameters:

U - the date type for the result

Parameters:

tf - the TensorFlow Ops encapsulating a Graph environment.

resultType - the data type for the result

Returns:

the result, possibly with control dependencies