Main Page | Modules | Namespace List | Class Hierarchy | Alphabetical List | Class List | Directories | File List | Namespace Members | Class Members | File Members | Related Pages

WinnerTakeAllTempNote Class Reference

#include <Neuro/WinnerTakeAll.H>

Inheritance diagram for WinnerTakeAllTempNote:

Inheritance graph
[legend]
Collaboration diagram for WinnerTakeAllTempNote:

Collaboration graph
[legend]
List of all members.

Detailed Description

Winner take all adapted for temporal noticing.

This is a trivial winner-take-all implementation, based on a 2D layer of LeakyIntFire neurons and a unique global inhibitory neuron. All neurons in the layer charge up in parallel; whenever one reaches threshold, it activates the inhibitory interneuron (which is connected to all neurons in the layer), which in turn resets all neurons in the layer. WinnerTakeAllStd is an Image<LeakyIntFire> and inherits all the standard methods of class Image. To avoid confusion, we here add explicit input() and getV() methods (rather than providing conversion functions between LeakyIntFire and float, which could make the process more transparent but also possibly more confusing).

Definition at line 350 of file WinnerTakeAll.H.

Public Member Functions

 WinnerTakeAllTempNote (OptionManager &mgr, const std::string &descrName="Winner-Take-All Temporal Noticing", const std::string &tagName="WinnerTakeAllTempNote")
 Ininitialized constructor.
virtual ~WinnerTakeAllTempNote ()
 Destructor.
virtual void reset1 ()
 Reset to initial state just after construction.

Protected Member Functions

virtual void input (const Image< float > &in)
 Set new input currents for all neurons.
virtual Image< float > getV () const
 Return all our membrane potential voltages as an Image<float>.
virtual void saccadicSuppression (const bool on)
 Turn saccadic suppression on/off.
virtual void blinkSuppression (const bool on)
 Turn blink suppression on/off.
Image< float > getVth (const bool normalize=false) const
virtual void integrate (const SimTime &t, Point2D< int > &winner)
 Integrate inputs until time t (in s) and update membrane potentials.
void inhibit ()
virtual void save1 (const ModelComponentSaveInfo &sinfo)
 Save our various results.

Protected Attributes

Image< LeakyIntFireAdpitsNeurons
segmentImageMC2< float, unsigned
int, 1 > 
itsMaskSegment
 Use the color segmenter to find a good mask region around winner.
std::vector< float > itsLowMaskBound
std::vector< float > itsHighMaskBound
Image< float > itsInitMask
LeakyIntFire itsGIN
SimTime itsT
float itsGleak
float itsGinh
float itsGinput


Constructor & Destructor Documentation

WinnerTakeAllTempNote::WinnerTakeAllTempNote OptionManager mgr,
const std::string descrName = "Winner-Take-All Temporal Noticing",
const std::string tagName = "WinnerTakeAllTempNote"
 

Ininitialized constructor.

The WTA will be resized and initialized the first time input() is called

Definition at line 561 of file WinnerTakeAll.C.

References GVX_TRACE, itsGIN, itsGleak, itsHighMaskBound, itsLowMaskBound, and LeakyIntFire::setGleak().

WinnerTakeAllTempNote::~WinnerTakeAllTempNote  )  [virtual]
 

Destructor.

Definition at line 575 of file WinnerTakeAll.C.

References GVX_TRACE.


Member Function Documentation

void WinnerTakeAllTempNote::blinkSuppression const bool  on  )  [protected, virtual]
 

Turn blink suppression on/off.

Reimplemented from WinnerTakeAllStd.

Definition at line 765 of file WinnerTakeAll.C.

References OModelParam< T >::getVal(), GVX_TRACE, inhibit(), WinnerTakeAllAdapter::itsUseBlinkSuppression, and LINFO.

Image< float > WinnerTakeAllTempNote::getV  )  const [protected, virtual]
 

Return all our membrane potential voltages as an Image<float>.

The curent values are possibly normalized to 0..255. The normalization will use itsSalmapFactor if it is non-null

Reimplemented from WinnerTakeAllStd.

Definition at line 710 of file WinnerTakeAll.C.

References Image< T >::begin(), Image< T >::beginw(), Image< T >::endw(), Image< T >::getDims(), GVX_TRACE, itsNeurons, NO_INIT, and ModelComponent::stop().

void WinnerTakeAllTempNote::input const Image< float > &  in  )  [protected, virtual]
 

Set new input currents for all neurons.

This will initialize and resize the network if the network is currently uninitialized (e.g., just after construction or reset()).

Reimplemented from WinnerTakeAllStd.

Definition at line 594 of file WinnerTakeAll.C.

References Image< T >::beginw(), Image< T >::endw(), Image< T >::getDims(), Image< T >::getHeight(), Image< T >::getWidth(), GVX_TRACE, height, Image< T >::initialized(), itsGinh, itsGleak, itsHighMaskBound, itsInitMask, itsLowMaskBound, itsMaskSegment, itsNeurons, Image< T >::resize(), segmentImageMC2::SIsetAvg(), segmentImageMC2::SIsetFrame(), segmentImageMC2::SIsetValThresh(), segmentImageMC2::SItoggleCandidateBandPass(), ModelComponent::stop(), width, and ZEROS.

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

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

Reimplemented from WinnerTakeAllStd.

Definition at line 624 of file WinnerTakeAll.C.

References Image< T >::begin(), Image< T >::beginw(), SimTime::computeDeltaT(), Image< T >::getHeight(), LeakyIntFire::getTimeStep(), LeakyIntFireAdp::getV(), Image< T >::getVal(), LeakyIntFireAdp::getVfire(), Image< T >::getWidth(), GVX_TRACE, h, i, Point2D< T >::i, inhibit(), LeakyIntFire::integrate(), itsGIN, itsGinput, itsGleak, itsInitMask, WinnerTakeAllAdapter::itsInputCopy, itsMaskSegment, itsNeurons, itsT, Point2D< T >::j, LINFO, LeakyIntFire::setG(), segmentImageMC2::SIreturnBlobs(), segmentImageMC2::SIreturnNormalizedCandidates(), segmentImageMC2::SIsegment(), sqrt(), and w.

void WinnerTakeAllTempNote::reset1  )  [virtual]
 

Reset to initial state just after construction.

Reimplemented from WinnerTakeAllStd.

Definition at line 581 of file WinnerTakeAll.C.

References Image< T >::freeMem(), GVX_TRACE, itsGinh, itsGinput, itsGleak, itsNeurons, itsT, WinnerTakeAllStd::reset1(), and SimTime::ZERO().

void WinnerTakeAllTempNote::saccadicSuppression const bool  on  )  [protected, virtual]
 

Turn saccadic suppression on/off.

Reimplemented from WinnerTakeAllStd.

Definition at line 756 of file WinnerTakeAll.C.

References OModelParam< T >::getVal(), GVX_TRACE, inhibit(), WinnerTakeAllAdapter::itsUseSaccadicSuppression, and LINFO.

void WinnerTakeAllTempNote::save1 const ModelComponentSaveInfo sinfo  )  [protected, virtual]
 

Save our various results.

Reimplemented from ModelComponent.

Definition at line 774 of file WinnerTakeAll.C.

References FLOAT_NORM_0_255, FLOAT_NORM_PRESERVE, getVth(), GVX_TRACE, itsInitMask, and SRC_POS.


Member Data Documentation

segmentImageMC2<float, unsigned int, 1> WinnerTakeAllTempNote::itsMaskSegment [protected]
 

Use the color segmenter to find a good mask region around winner.

Definition at line 390 of file WinnerTakeAll.H.

Referenced by input(), and integrate().


The documentation for this class was generated from the following files:
Generated on Mon Nov 23 15:51:29 2009 for iLab Neuromorphic Vision Toolkit by  doxygen 1.4.4