|| || || || || || || ||

---

Todo List

Member Impala::Core::Training::ApplyConceptsHelperKernels::Predict (String modelName)

perhaps we can use the code in SvmHelper?

Class Impala::Core::Table::CriterionElement1InSet< T >

make a faster implementation, this one's O(n^2), can at least be O(n*log(n)), can be O(n) when both tables are sorted

Class Impala::Core::DataFactory

Right now this class assumes a video set. We should find a way to generalise to RawDataSet, although perhaps those classes will be refactored anyway.

Member Impala::Core::DataFactory::MakeImage (Quid q)

check that this is really the idiomatic way to get an image

Member Impala::Core::DataFactory::WriteConceptModel (String concept, Training::Classifier *c)

make a function SVM::SaveModel(IOBuffer*) and change this function accordingly

Member Impala::Core::Array::Trait::FuncKalman::doIt (const Vec3Real64 &p1, const Vec3Real64 &p2, int x, int y)

check correctness of outliers during occlusion

Member Impala::Core::ImageSet::ImageSet::GetImage (int fileId)

GetImageData is experimental, if/when it becomes standard This function should use the output of GetImageData and just do a ReadImageFromMemory on it.

Member Impala::Core::Array::Trait::InOutSetRandom::DoIt ()

change implementation

Class Impala::Core::VideoSet::LbpEval

refactor this -> LpbEval should be split, there is duplication of code in ProtoSimilarityEval. Make a baseclass for Feature::Computor and derive an Lbp computor thereof.

Member Impala::Core::Training::LogisticRegression::Train (PropertySet *properties)

replace by 'minus infinity'

Member Impala::Core::Training::LogisticRegression::Train (PropertySet *properties)

replace by 'real max'

Member Impala::Core::Training::LogisticRegression::Train (PropertySet *properties)

replace by 'real min'

Member Impala::Visualization::Plot::Plottable::Draw (Plot *plot)=0

decide what coordinate system to draw in, for now assume [-1..1] in all dimensions

Class Impala::Core::Training::Svm

Fix the machanism to set gamma to 1/vectorlength. See OverrideModelOptions(), Train() and MakeSvmParams().

Member Impala::Core::Training::SvmProblemBuilder::AddFeatureTable (Feature::FeatureTable *table)

We can speed things up by assuming everything is sorted on quids, in that case we don't have to search the QuidTable with GetIndex

Class Impala::Core::Tracking::TrackerBackFore

add parameters (such as sigma, number of orientations) to constructor (perhaps in struct)

Member Impala::Core::Tracking::TrackerBackFore::FillSearchArea (Array::Array2dVec3Real64 *searchArea, int bw, int bh)

PatSet in such a way the border is filled with a part of the image in stead of this ugly hack

Member Impala::Core::Tracking::TrackerBackFore::FillSearchArea (Array::Array2dVec3Real64 *searchArea, int bw, int bh)

introduce this hack as a temporary pattern, this make it possible to reason about the code

Member Impala::Core::Tracking::TrackerGeneralised::Clip (Position &pos)

take rotation and scale into consideration.

Member Impala::Application::GetPartial (Feature::FeatureTable *f, int partnumber, int partcount)

find better name

Member Impala::Core::Array::InvQuasi (Array2dVec3Real64 *src, double sigma, double precision)

perhaps this function should be split in 3 functions that compute the quasi invariant and variant for the three categories...

Member Impala::Core::Training::LoadClassifier (CString modelFileName, Util::Database *db)

:

• read model and find out what classifier it belongs to
• all classifiers have a constructor that takes a models filename as argument

Member Impala::Core::Vector::Apply (double(*fn)(const VectorTem< double > &, const VectorTem< double > &), const VectorSet< Array::Array2dScalarReal64 > *horizontal, const VectorSet< Array::Array2dScalarReal64 > *vertical)

apply to pattern coding style, introduce function object i.s.o. function pointer

Member Impala::Core::Vector::HistogramAccumulation (VectorSet< ArrayT > *vs, bool doNormalizeHeight, bool doNormalizeWeight, bool *filter)

: the name is not correct??

Member Impala::Util::Broadcast (IOBuffer *buffer)

investigate performance of MPI vs, Franks implementations

Namespace OglGui

XXX RvB: apart from SetAllowSizeDirection, also implement function to disable a certain direction, leaving other settings alone: SetDisableSizeDirection. Same goes for move: SetDisableMoveDirection. Or alternatively add a flag to SetAllowSizeDirection(dir, bool mode) latter method is preferred

---