WinnerTakeAllGreedy Class Reference

A greedy attention selection mechanism, not exactly winner-take-all. More...

#include <Neuro/WinnerTakeAll.H>

Inheritance diagram for WinnerTakeAllGreedy:

[legend]

Collaboration diagram for WinnerTakeAllGreedy:

[legend]

List of all members.

Public Member Functions
	WinnerTakeAllGreedy (OptionManager &mgr, const std::string &descrName="Winner-Take-All Greedy", const std::string &tagName="WinnerTakeAllGreedy")
	Ininitialized constructor.
virtual	~WinnerTakeAllGreedy ()
	Destructor.
Protected Member Functions
virtual void	integrate (const SimTime &t, Point2D< int > &winner)
	Integrate inputs until time t (in s) and update membrane potentials.
Protected Attributes
OModelParam< float >	itsThresholdFac
	Threshold factor, in [0.0 .. 1.0].

---

Detailed Description

A greedy attention selection mechanism, not exactly winner-take-all.

This will select, among a set of possible candidate targets that are above a threshold, the one that is closest to current eye position. It hence relies on being passed a valid current eye position in the queue. Note that the returned winner could be exactly at our current eye position if that is above threshold; we here do not attempt to force it to be some distance away from current fixation, and instead rely on IOR in the saliency map to enforce that if desired.

Definition at line 315 of file WinnerTakeAll.H.

---

Constructor & Destructor Documentation

WinnerTakeAllGreedy::WinnerTakeAllGreedy	(	OptionManager &	mgr,
		const std::string &	descrName = "Winner-Take-All Greedy",
		const std::string &	tagName = "WinnerTakeAllGreedy"
	)

Ininitialized constructor.

Definition at line 451 of file WinnerTakeAll.C.

WinnerTakeAllGreedy::~WinnerTakeAllGreedy

(

)

[virtual]

Destructor.

Definition at line 461 of file WinnerTakeAll.C.

---

Member Function Documentation

void WinnerTakeAllGreedy::integrate	(	const SimTime &	t,
		Point2D< int > &	winner
	)			[protected, virtual]

Integrate inputs until time t (in s) and update membrane potentials.

We operate like a WinnerTakeAllStd, except that when a winner is found, we look around for locations that are above threshold, and pick the closest one to our current eye position.

Reimplemented from WinnerTakeAllStd.

Definition at line 467 of file WinnerTakeAll.C.

References Image< T >::begin(), Image< T >::beginw(), chamfer34(), SimTime::computeDeltaT(), Image< T >::coordsOk(), eye(), Image< T >::getDims(), Image< T >::getHeight(), LeakyIntFire::getTimeStep(), WinnerTakeAllStd::getV(), Image< T >::getVal(), OModelParam< T >::getVal(), Image< T >::getWidth(), Point2D< T >::i, LeakyIntFire::integrate(), isLocalMax(), Point2D< T >::isValid(), WinnerTakeAllAdapter::itsEyePos, WinnerTakeAllAdapter::itsInputCopy, WinnerTakeAllAdapter::itsLevelSpec, itsThresholdFac, LevelSpec::mapLevel(), LeakyIntFire::setG(), and ZEROS.

---

Member Data Documentation

OModelParam<float> WinnerTakeAllGreedy::itsThresholdFac [protected]

Threshold factor, in [0.0 .. 1.0].

Definition at line 333 of file WinnerTakeAll.H.

Referenced by integrate().

---

The documentation for this class was generated from the following files:

• src/Neuro/WinnerTakeAll.H
• src/Neuro/WinnerTakeAll.C