Member drwnAsyncMaxProdInference::decodeBeliefs (drwnFullAssignment &mapAssignment)

improve this? consisted decoding?

Member drwnColourHistogram::accumulate (unsigned char red, unsigned char green, unsigned char blue)

interpolate between 8 neighbouring bins

Member drwnCreateHeatMap (const cv::Mat &m, drwnColorMap cm=DRWN_COLORMAP_RAINBOW)

refactor to use drwnCreateHeatMap(m, vector<CvScalar>) variant

Member drwnDataset< XType, YType, WType >::subSample (int sampleRate, bool bBalanced=false)

consider sample weighting

enforce balanced for discrete targets only

Member drwnDualDecompositionInference::inference (drwnFullAssignment &mapAssignment)

change for min-marginal subgradient

Class drwnFactor

move into own file.

allow for more efficient storage of higher-order factors

move to a templated design

Class drwnFactorGraph

template on a particular factor type (could be all table factors, for example, or generic drwnFactors)

Member drwnFactorGraphUtils::removeUniformFactors (drwnFactorGraph &graph)

reduce factors where variable cardinality is one

Member drwnFactorReduceOp::execute ()

speed up

Member drwnGEMPLPInference::inference (drwnFullAssignment &mapAssignment)

add this to computation graph

Member drwnHistogram< TYPE >::mergeWeightedHistogram (const drwnHistogram &H, double w)

needs to be extended

Class drwnInference

derive from drwnStdObjectIface to allow factory creation

Member drwnLPSolver::solve ()

solve with blockwise elimination

Class drwnMAPInference

derive from drwnStdObjectIface to allow factory creation

Class drwnMAPInferenceFactory

use drwnFactory templated class (but then drwnMAPInference needs to be default constructable)

Member drwnMaskedPatchMatch::visualize () const

Add modifySourceImage and modifyTargetImage functions based on a mask

Member drwnMaxProdInference::decodeBeliefs (drwnFullAssignment &mapAssignment)

improve this? consisted decoding?

Member drwnMergeSuperpixels (const cv::Mat &img, cv::Mat &seg, unsigned maxSegs)

look for "best matching" neighbour to merge with

Member drwnNNGraphMoves::rescore (drwnNNGraph &graph, unsigned imgIndx, const DistanceMetric &M)

delete the node

Class drwnNNGraphMoveUpdateThread< DistanceMetric >

features are replicated for positive and negative nn graphs which is a memory waste. Try refactor with features external.

Member drwnPatchMatchGraphLearner::appendCIELabFeatures (const cv::Mat &img, cv::Mat &features, int nChannel=0) const

move these elsewhere

Member drwnPatchMatchGraphLearner::cacheImageFeatures ()

allow a pixel feature generator to be provided caches features for all image pyramids

Member drwnPatchMatchGraphLearner::initialize ()

cache image features here instead of in constructor?

Class drwnPatchMatchImagePyramid

make cacheable so drwnPatchMatchGraph can handle thousands of images

Class drwnPersistentStorageBuffer< T >

rename to drwnPersistentRecordBuffer

Class drwnPixelSegCRFInference

add custom addAuxiliaryTerms function

Member drwnPixelSegModel::cacheBoostedPixelResponses (drwnSegImageInstance &instance) const

refactor using persistent storage class

Class drwnProperties

Could also store within property interface object (which would make for slower access within class members, but easier declaration of lots of properties).

Member drwnQPSolver::lineSearchGeneral (const VectorXd &x, const VectorXd &dx, const VectorXd &nu, const VectorXd &dnu) const

implement

Member drwnQPSolver::solveSimplex ()

implement specialized simplex solver

Member drwnSLICSuperpixels (const cv::Mat &img, unsigned nClusters, double spatialWeight=200.0, double threshold=1.0e-3)

replace with separate x and y grid sizes

Class drwnSparseLPSolver

refactor drwnLPSolver code to use templated sparse/dense matrix

Member drwnSparseLPSolver::solve ()

solve with blockwise elimination

Class drwnTableFactor

drwnSparseTableFactor class and drwnPottsFactor class

Class drwnTableFactorMapping

create a multi-table mapping (will reduce amount of repeated computation in iterators)

specialize for unary and pairwise cases

Member drwnTableFactorMapping::drwnTableFactorMapping (const vector< int > &dstVars, const vector< int > &srcVars, const drwnVarUniversePtr &pUniverse)

group variables as above if appear contiguously in src and dst

Member drwnTLinearRegressor< FeatureMap >::getRegression (const vector< double > &features) const

define feature mapping functions in terms of Eigen::VectorXd

Member drwnTMultiClassLogistic< FeatureMap >::getClassScores (const vector< double > &features, vector< double > &outputScores) const

define feature mapping functions in terms of Eigen::VectorXd