Package org.ddogleg.optimization.loss

Class LossCauchy

java.lang.Object

org.ddogleg.optimization.loss.LossFunctionBase

org.ddogleg.optimization.loss.LossCauchy

Direct Known Subclasses:

LossCauchy.Function, LossCauchy.Gradient

public abstract class LossCauchy
extends LossFunctionBase

Loss function inspired by the Cauchy distribution, a.k.a Lorentzian loss function.

L(x) = a2(log(1+(x/a)2)

, where 'x' is the residual and 'a' is a scale parameter. Smaller values of the tuning parameter 'a' make it behave more similar to the L2 loss while larger values make it more robust to outliers.

1. Black, Michael J., and Paul Anandan. "The robust estimation of multiple motions: Parametric and piecewise-smooth flow fields." Computer vision and image understanding 63.1 (1996): 75-104.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	LossCauchy.Function	Implementation of the smooth Cauchy loss function
static class	LossCauchy.Gradient	Implementation of the smooth Cauchy loss gradient

Field Summary

Fields

Modifier and Type	Field	Description
protected double	alpha

Fields inherited from class org.ddogleg.optimization.loss.LossFunctionBase

numberOfFunctions

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	LossCauchy(double alpha)

Method Summary

Methods inherited from class java.lang.Object

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

Field Details

alpha

protected double alpha

Constructor Details

LossCauchy

protected LossCauchy(double alpha)

Parameters:

alpha - scale parameter that changes sensitivity to outliers