Minos (MinosCVC.dll) 15.0
TLearnControlStructure Struct Reference

Parameter collection for the generation of Minos classifiers. More...

Data Fields

void * Data
 Pointer to the user's private data to be used for progress reporting. More...
 
cvbval_t Param1
 Indifference Radius. Recommended value is 6. More...
 
cvbval_t Param2
 Negative Density. Recommended value is 1000. More...
 
cvbval_t Param3
 Minimum feature contrast. Recommended value is 8 More...
 
cvbval_t Param4
 Undisclosed parameter. Always set to 15.
 
cvbval_t Param5
 Undisclosed parameter. Always set to 2.
 
cvbval_t Param6
 Minimum feature count. Recommended value is 20 More...
 
cvbval_t Reserved [92]
 Reserved for future use.
 
TFLearn ShowProgress
 Progress reporting callback. More...
 

Detailed Description

Parameter collection for the generation of Minos classifiers.

This data structure is used and evaluated by the functions LearnCLFFromMTS and LearnCLFFromMTSW. It contains all the parameters that Minos needs to take into account for classifier generation.

Note that the parameters have initially been given names that do not make their purpose apparent. For backward compatibility this has not been changed later on, so please make sure to read the parameter documentation carefully when using this structure.

Field Documentation

◆ Data

void* Data

Pointer to the user's private data to be used for progress reporting.

The value set here will be passed to the TFLearn callback whenever it is invoked during the learning process.

◆ Param1

cvbval_t Param1

Indifference Radius. Recommended value is 6.

The indifference radius defines squares whose midpoint lies at the anchor point of each positive instance. Points inside this square are never considered to be negative instances during the learning process. In other words this parameter defines the extent of a set of regions that may be described as "a no-man's land" whose content neither clearly belongs to the pattern class nor to the excluded class.

Usually the default value of 6 is going to yield reasonable results. If all the models in a training set are exceptionally big (roughly 200x200 pixels or more) it might make sense to slightly increase this value to prevent the classifier from becoming too strict. If the models are very small (only of the order of 15x15 pixels or less) it might make sense to slightly reduce this value.

◆ Param2

cvbval_t Param2

Negative Density. Recommended value is 1000.

The negative density defines the scan density Minos should use when checking a training set image for negative instances, counter samples for all the trained models. Valid settings are in the range [0...1000]. Higher values tend to increase the classifier quality as well as (only slightly!) the time required for classifier generation. Low values may increase the risk of the resulting classifier for detecting false positives. It is generally recommended to use a value of 1000 here.

◆ Param3

cvbval_t Param3

Minimum feature contrast. Recommended value is 8

Minos uses contrast based features, and the minimum contrast required for a feature may be defined with this member. Usually the recommended value of 8 yields reasonable results, but in extreme low contrast situations it might make sense to reduce this value.

◆ Param6

cvbval_t Param6

Minimum feature count. Recommended value is 20

With this parameter Minos can be forced to build bigger classifiers that may in some fringe cases be more robust. Usually 20 features per model are sufficient, but in some cases (training sets with a very high number of models and/or many very similar models) a classifier with a higher feature count may yield better classification results (at the expense of increased processing time).

◆ ShowProgress

TFLearn ShowProgress

Progress reporting callback.

This callback will be invoked during the learning process after every processed training set instance. If you do not need or want to use progress reporting set this value to nullptr.