deep_surface_svl_simple.c File Reference

Tutorial 13 - Deep Surface - How to create a project, import defects labeling from images, and execute the training (SVL). More...

#include "../common/common.h"

Include dependency graph for deep_surface_svl_simple.c:

Go to the source code of this file.

Data Structures
struct	sb_svl_user_data

Functions
sb_t_err	create_deep_surface_project_file (SB_HANDLE *deep_surface)
	Create a solution file with a Deep Surface project. More...
sb_t_err	set_labeling (SB_HANDLE deep_surface)
	Set labeling. More...
sb_t_err	execute_training (void)
	Execute training. More...
sb_t_err	check_labeling_file_exists (char *img_file, sb_t_folder *folder_defects, int *file_index)
	Check if a labeling mask file exists for the specified image. More...
int	main (void)

Detailed Description

Tutorial 13 - Deep Surface - How to create a project, import defects labeling from images, and execute the training (SVL).

Attention

To execute this tutorial you need:

• to have installed the SB Deep Learning Framework
• to ensure that path to SB Deep Learning Framework is included in the PATH enviroment variable (LD_LIBRARY_PATH in Linux) or in any case to be visible by the program. In Windows it is possible to specify additional paths as argument to the sb_init_dl function. as descripted in the manual at Appendix 2: configure tutorials to use SB Deep Learning Framework with Qt Creator in Linux .

This tutorial shows what you should do to train a model from scratch, i.e. without using the SB GUI, starting from a set of images with associated labeling masks. A labeling mask is a grayscale image containing roi defects. On the command line you can specify the memory percentage to leave free.
During the execution the programs shows the status of the svl with the progressive results.
If you use Microsoft Visual Studio, you can stop the SVL using a key combination.
At the end the new SVL is saved in a solution file.

Attention

In current directory you should copy the license file sb.lic.
Alternatively, if you have bought a the dongle license, you can connect it to an usb port of the PC.

See also

sb_init

sb_init_dl

sb_release

sb_get_info

sb_license_get_info

sb_folder_load

sb_folder_destroy

sb_solution_get_info

sb_solution_destroy_info

sb_project_load

sb_project_destroy

sb_project_save

sb_project_get_par

sb_project_set_par

sb_par_format

sb_svl_reset

sb_svl_stop_request

sb_svl_run

sb_svl_get_res

sb_svl_destroy_res

Definition in file deep_surface_svl_simple.c.

Function Documentation

check_labeling_file_exists()

sb_t_err check_labeling_file_exists	(	char *	img_file,
		sb_t_folder *	folder_defects,
		int *	file_index
	)

Check if a labeling mask file exists for the specified image.

1. Get the name without extension of the input image file.
2. Scan the folder with the defects labeling and check if a mask with name equal to the current image exists.
3. If the defects mask is founded, the index of the file in the folder is assigned.

Definition at line 210 of file deep_surface_svl_simple.c.

Here is the call graph for this function:

create_deep_surface_project_file()

sb_t_err create_deep_surface_project_file

(

SB_HANDLE *

deep_surface

)

Create a solution file with a Deep Surface project.

1. Create the project file
2. Set parameters of the model of the object.
   • Set the model name
3. Save the solution file.

Definition at line 188 of file deep_surface_svl_simple.c.

Here is the call graph for this function:

execute_training()

sb_t_err execute_training

(

void

)

Execute training.

1. Get the solution information.
2. Load the project.
3. Reset a previous training.
4. Set training parameters.
   • set GPU device for SVL
   • set deep learning network parameters (SDINet)
   • set deep learning perturbation parameters (flip, rotation and shift)
5. Start the SVL (only on Linux)
6. Save SVL results if it is possible
7. Release the handles and the library

Definition at line 309 of file deep_surface_svl_simple.c.

Here is the call graph for this function:

main()

int main

(

void

)

In the following the list of the program steps:

1. Initialization of the SqueezeBrains library.
2. Initialization of the Squeezebrains deep learning modules.
3. Wait until the license is active.
4. Create SB project file.
5. Add defects to images from labeling masks.
6. Execute training.
7. Destroy the project handle.
8. Release the library.

Definition at line 137 of file deep_surface_svl_simple.c.

Here is the call graph for this function:

set_labeling()

sb_t_err set_labeling

(

SB_HANDLE

deep_surface

)

Set labeling.

1. Erase the old rtn files to begin from zero
2. Load folder containing the labeling masks
3. Add the labeling to the images. For each image:
   • Set the flag classified of the image in order for the image to be considered for the learning
   • Set the roi
   • Set the roi defects loading data from the corresponding labeling mask
4. Destroy the handles

Definition at line 235 of file deep_surface_svl_simple.c.

Here is the call graph for this function: