drwnNNGraphProjectFeaturesThread Class Reference

Inheritance diagram for drwnNNGraphProjectFeaturesThread:

Public Member Functions
	drwnNNGraphProjectFeaturesThread (unsigned imgIndx, const drwnNNGraph &srcGraph, drwnNNGraph &posGraph, drwnNNGraph &negGraph, const MatrixXf &Lt)
	drwnNNGraphProjectFeaturesThread (const set< unsigned > &imgIndxes, const drwnNNGraph &srcGraph, drwnNNGraph &posGraph, drwnNNGraph &negGraph, const MatrixXf &Lt)
void	operator() ()
	thread functor called by drwnThreadPool with the appropriate threadId More...

Protected Attributes
set< unsigned >	_imgIndxes
const drwnNNGraph &	_srcGraph
drwnNNGraph &	_posGraph
drwnNNGraph &	_negGraph
const MatrixXf &	_Lt

Static Protected Attributes
static map< unsigned, MatrixXf >	_imgFeatureData

Additional Inherited Members
Protected Member Functions inherited from drwnThreadJob
unsigned	threadId ()
	obtain the id for the thread running (or that ran) this job. The return value is guaranteed to be between 0 and numThreads. Useful for accessing a global resource without colliding with other jobs.
void	lock ()
	acquire a lock (on all jobs in the same thread pool)
void	unlock ()
	release the lock (on all jobs in the same thread pool)

Member Function Documentation

void drwnNNGraphProjectFeaturesThread::operator() ( )

inlinevirtual

thread functor called by drwnThreadPool with the appropriate threadId

Performing feature transformation on a block results is about a factor of two speedup. This suggests moving feature storage from nodes to images. However, the change will make introduction of different node types (with different sized feature vectors) difficult. It also means that the metric functors would need to be passed a copy of the graph in addition to the nodes.

Implements drwnThreadJob.

---

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

• drwnNNGraphLearn.cpp