SqueezeBrains SDK 1.13
sb_t_par_sl Struct Reference

Shallow Learning modules parameters. More...

#include <sb.h>

Data Fields

float speed_boost
 Detection speed boost. More...
 
int detection_out_of_roi
 Allow the detection of occurrencies at the extremities or partially outside the analysis roi. More...
 

Detailed Description

Shallow Learning modules parameters.

Definition at line 10600 of file sb.h.

Field Documentation

◆ detection_out_of_roi

int sb_t_par_sl::detection_out_of_roi

Allow the detection of occurrencies at the extremities or partially outside the analysis roi.

The parameter changes the results of the function sb_project_detection and has a different effect whether the project is of Retina or Surface type.
For a Surface project the parameter enable the calculation of the voting plan on the entire analysis ROI, so it is possibile to detect defects that are near the border of the analysis ROI.
In the image below an example of a defect near the border of the analysis ROI detected by sb_project_detection . In this case the defect is also near the border of the image because the analysis ROI is extended to the whole image.

Detection of an defect near the border of the analysis ROI

For a Retina project the parameter enables the search of objects are partially outside the roi.

Detection of an object partially out of the roi

Used only by Retina and Surface projects.

Definition at line 10639 of file sb.h.

◆ speed_boost

float sb_t_par_sl::speed_boost

Detection speed boost.

The parameter allows the user to increase the elaboration speed (i.e. to reduce the execution time), of the sb_project_detection function.
The effectiveness of the parameter depends on how the training was done. Usually, at least 10 samples per model are required so that the SVL correctly sets the parameters necessary to perform the speed boost. The parameter does not always have the same effectiveness and under certain conditions it can even increase the calculation time of the sb_project_detection function. For example, if the size of the ROI of the analysis is comparable to the size of the sample, the speed boost increase the time.
The speed boost works as follows: when the model window scans the analysis ROI, instead of doing all the calculations to evaluate the weight of the window in a given position, the calculations are divided by stages and if a stage does not reach a certain score, them the function will jump to the next position.
The range of values is between 0 and 1, where 0 means speed boost disabled and 1 means maximum speed boost.
Usually there is already a great reduction over time already at a value of 0.1 and therefore, as the value increases, the reduction is less significant.
Used only by Retina project.

Attention
  • The higher the value the higher is the possibility of generating FALSE NEGATIVE.
  • Set speed boost to a value > 0 is not compatible with a value of the parameter sb_t_par_model::num_occurrences different from 0.
    True Negative samples may not be exported.
  • If speed boost is enabled, the results obtained by the SVL and those obtained in the test may be different due to the False Negative samples generated by the speed boost.
See also
License configurations

Definition at line 10625 of file sb.h.


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