- Member AdaBoost< BoostableClassifier< WeakClassifier > >::Update (BoostableClassifier< WeakClassifier > &H, DataSetType &training_set, DataSetType &validation_set, bool verbose=true)
- qui c'e' un problema di fondo. Ovvero alpha lo calcola AdaBoost massimizzandolo sul training set. Tuttavia in MAdaBoost alpha e' calcolato con un algoritmo diverso da quello di AdaBoost in quando il training set ha un peso differente dal set che viene passato a questa funzione. Alpha va calcolato sul training_set, e aggiornato sul validation
- Member BoostClassifier< Classifier >::sort ()
- check)
- Member ClassifierDetectHelper< IntegralChannelImagePreprocessor, _Instance >::scale_object_detector (std::vector< Candidate > *out, const ImageHandle *src, double s, int s_index)
missing m_searchRoi
missing
- Class DecisionStump< DataType >
- try to compact BoostableClassifier with DecisionStump in order to reduce number of multiplication (parity * alpha) (e.g. using RealDecisionStump classes)
- Member frameidx
- remove it
- Member IntegralImageHandle< T >::Export (D *buf, int x0, int y0, int dst_width, int dst_height) const
- handle cropping better
- Member PrecomputedPatternResponse< ResponseType >::Precompute (const TrainintSetType &set, const std::vector< FeatureType > &feats, int max_concurrent_jobs)
- multithreading!
- Member ResetWeight (DataSet &list, double priori_knownledge)
- deal with asymmetric adaboost and not use priori_knownledge
- Member SoftCascadeClassifier< T >::raw (const TParam1 data, const TParam2 p1) const
- make name homogeneous with others mechanism
