SearchPredictorFactory Properties

Range of angles to be covered during classifier training.

Radius for extracting positive search instances. The unit size is the size of the feature window, i.e. a value of 0.5 means that positive samples can be extracted from withing a range of +/- Feature Window Width and +/- Feature Window Height around each trained instance. Smaller values will necessitate a finer search grid but may yield better results on difficult search tasks.

Feature resolution (determines the size of the classification retina) for step 3 and up.

Feature resolution (determines the size of the classification retina) for search step 1 and 2.

Invariances to be used for generating the classifier.

Regularization value to be used for generating the object. Possible values range from 0 to 10, good starting values for experiments are usually around 0.01.

Number of classification steps during search operation.

Preprocessing code with which the object is to be generated.

Sample size that has been used in each training set.

Range of scale factors to be covered during classifier training. Only meaningful if the invariance type RotationScaleTranslation has been set.

Range of permissible singular values to be covered during classifier training. Only meaningful if the invariance type AffineGroup has been set.