SB SDK 1.11
sb_t_svl_sl_par Struct Reference

SVL parameters that configure the Shallow Learning training. More...

#include <sb.h>

Data Fields

float goodness_target
 Goodness target of the training. More...
 
char features [SB_PAR_FEATURES_NAMES_LEN]
 List of the features among which SVL will choose the best features. More...
 
sb_t_svl_par_optimization_mode optimization_mode
 Optimization mode. More...
 
int auto_levels
 Enable the automatic Surface levels training. More...
 

Detailed Description

SVL parameters that configure the Shallow Learning training.

Used only by Retina and Surface projects.

Definition at line 9098 of file sb.h.

Field Documentation

◆ auto_levels

int sb_t_svl_sl_par::auto_levels

Enable the automatic Surface levels training.

When this flag is enabled, the SVL estimates the best levels to be trained and at the end of the training set the selected scale levels into the sb_t_par structure.
The SVL estimates the best levels by analyzing the labeling of defects and in particular, the most important characteristic of defects is their size. The choice of levels is made only when training from scratch, so it is advisable to reset it if in labeling new images, or in retouching existing labeling, the minimum size or maximum size is significantly changed.
Used only by Surface projects.

Note
If enabled you should not modify the sb_t_par_model::levels list of the models.
See also
Levels
sb_t_par_levels

Definition at line 9151 of file sb.h.

◆ features

char sb_t_svl_sl_par::features[SB_PAR_FEATURES_NAMES_LEN]

List of the features among which SVL will choose the best features.

The features are separated by the SB_DELIMITER character.
Used only by Retina and Surface projects.

Features selection
See also
Features
License configurations

Definition at line 9126 of file sb.h.

◆ goodness_target

float sb_t_svl_sl_par::goodness_target

Goodness target of the training.

The goodness is the separation between the weight or confidence of TRUE POSITIVE and TRUE NEGATIVE samples, i.e. between the foreground and the background, or, in case of Surface project, between good and defect. The training tries to reach the target and stops when it reaches this value otherwise it stop before.
If the training takes a long time and the goodness is already good, you can also manually stop the operation.
Note that the value influences also the over / under fitting and the training time.
The range of values is between 0 and 1.
The default value is 0.5.
Used only by Retina and Surface projects.

See also
SVL goodness
Under and Over fitting
Long training times
sb_svl_stop_request

Definition at line 9116 of file sb.h.

◆ optimization_mode

sb_t_svl_par_optimization_mode sb_t_svl_sl_par::optimization_mode

Optimization mode.

The set of features effectively used for the training varies according to the optimization mode parameter and to the type of project.
See sb_t_svl_par_optimization_mode for more information.
Used only by Retina and Surface projects.

Training optimization mode
See also
sb_t_svl_par_optimization_mode
SVL - training
Features

Definition at line 9138 of file sb.h.


The documentation for this struct was generated from the following file: