The module name is used to identify it. By default, the name is set to the current date and time.
Some modules allow you to train an already trained model further. If selected, the network type selected will be identical to the trained one. If the model uses the view-finder, it will also be automatically set in compliance with the trained one.
This augments the images that go into training. It makes generalization of the model easier. However, it is not suitable for all cases.
Width shift (min 0, max 100)
A shift to the left or right will occur.
Height shift (min 0, max 100)
A shift up or down will occur.
The network type determines network accuracy and the amount of time necessary for the image to be evaluated. The faster the network, the less accurate it is.
Color jittering (min 0, max 50)
Color jittering, suitable for cases where images have different colors or shades.
Number of training cycles (min 1, max 10000)
The more training cycles, the less general the network is. If the differences between OK and NOK images are not very significant, more cycles are better. We recommend testing more training options and evaluating which model gives the best results.
Resistance to deviation (min 0, max 100)
Can be used to ignore noise or other deviation in the image.
Brightness resistance (min 0, max 100)
Resistance to changes in lighting.
Shear (min 0, max 360)
Object can be skewed by given factor in the direction of x or y axis.
Training begins by pressing the ‘Start training’ button. During the training, you can see a chart (not applicable for Anomaly of Surface and Surface Detection modules). Based on this chart, you need to stop the training at the right time. The chart should gradually decrease over the training duration. With decreasing, the resulting model is improving. Training can be ended at any time, preferably when the chart is not decreasing any more. Ideally, the chart should drop below at least one of the green lines. However, it is possible to stop the training early, test it and, if necessary, use the ‘Extend Model’ function to continue. The following chart shows the ideal position to stop the training.