Unsupervised Type Red Analyze Parameters

The parameters adjust how the tool is trained, and also how the tool processes statistical results. You can open and close the parameters side pane by selecting the tool and then clicking the % icon. You can access additional parameters that provide advanced features by selecting Expert Mode in the Help menu.

Type, Mode, and Network Model Parameters

Parameter Expert Mode Description
Type No Specifies the type of neural network model Each VisionPro Deep Learning tool is a neural network. A neural network mimics the way biological neurans work in the human brain. The neural network consists of interconnected layers of artifical neurons, called nodes, and they have multiple layers. Neural networks excel at tasks like image classification and pattern recognition..
Mode No

Specifies the tool mode, Legacy or Unsupervised.
Network Model No

Specifies the size of the neural network model A specific spatial arrangement of a set of features (Blue Locate and Blue Read tools only.) During a post-processing step, the Blue Locate and Blue Read tools can fit all of the features detected in an image to the models defined for the tool. The overall pose and identity of the model is then returned., which affects the time required for training Training is the process that your tool, which is a neural network, is learning about the features (pixels) based on the labels you made. For example, a tool will learn the defect/normal pixels in each image based on the defect/normal labels you drew. The goal of the tool Training is learning enough to give the correct inspection results of whether an unseen image is defective or not. The key to training is to ensure that you include all possible variations within your training set, and that your images are accurately labeled. Training times vary by the application, tool setup and the GPU in the PC being used to train the network. and processing. The Network Model parameter allows you to reduce the size of the trained tool network, which reduces the amount of time required for processing. The default value, Large, retains the standard tool behavior and performance. The two other values, Small and Medium, provide progressively faster runtime execution.

Selecting a value other than Large can result in different tool results. These differences are typically minor, however, when using this option, you need to validate tool performance using your own labeled data.

Sampling Parameters

The Sampling parameters control feature A feature is a visually distinguishable area in an image. Features typically represent something of interest for the application (a defect, an object, a particular component of an object). sampling during training and processing. For more information, see Feature Sampling.

Note: If you change the Sampling parameters of an already trained tool, you must retrain the tool because the underlying image statistics might be changed.
Parameter Expert Mode Description

Feature Size

 No

Specifies the typical feature diameter in pixels. The Feature Size is displayed in the lower left of the image, and you can resize it within the image to set a more accurate size.

The feature size The subjective size of the image features that you feel are most important for analyzing image content. The feature size determines the size of the image region used for sampling. Only used with Blue Locate, Blue Read, and Legacy type tools. should approximate the size of a typical defect. For elongated defects, such as streaks, smears, scratches and tears, the feature size should approximate the width of the defect.

1

Defect

2

Feature Size

Feature size strongly influences processing time (n2). In other words,, a Feature Size of 100 is 100 times faster than a size of 10, while a Feature Size less than 15 usually does not yield good results.

When setting the Feature Size, consider the following in regards to processing time (Ptime):

Note: The tool sees an area five times larger than the Feature Size setting. However, the level of detail is much higher in the center of the feature than in the periphery. For more information, see Feature Size.
Note: The Iterations parameter provide additional control over sampling at runtime. You can specify that a number of additional interpolated samples are collected in regions with detected defects. This can be useful for images where the range of complexity is highly variable within individual images.

Color

 No

Specifies the number of color channels to use when sampling the image. When set to 1, color images are converted to greyscale. The following values are available:

  1. Treats the image as greyscale
  2. Two channels images (spectral images, gray+alpha)
  3. BGR image.
  4. BGR(A) image
Note:
  • The VisionPro Deep Learning tools use the BGR channel order. RGB(A) images are converted to greyscale. For computational efficiency (memory allocation, transfer, file save, color conversion,… ), always prefer the correct number of channels.

  • If your application relies on color images, only use the minimum number of required color channels, and only send images that already have the correct number of channels to avoid conversion. This is because:

Number of Image Channels Number of Training Channels Description
1 (greyscale) 1 The correct setting for a greyscale image.
1 2, 3, 4 Most likely, this results in a training error.
2 1 The tool uses only the first channel.
2 2 The tool uses the full pixel information.
2 3, 4 This results in a training error.
3 (BGR) 1 This results in a BGR to greyscale conversion.
3 2 The tool only uses the first two channels (in other words, B and G).
3 3 The tool uses the full pixel information.
3 4 Most likely, this results in a training error.
4 (BGRA) 1 This results in a BGR(A) to greyscale conversion.
4 2 The tool only uses the first two channels (in other words, B and G).
4 3 The tool only uses the first three channels (in other words, B, G and R).
4 4 The tool uses the full pixel information.
Border Type Yes

Specifies how pixels on the outside of the image are sampled. Adding an additional mask at the boundaries of the image highly reduces the false detection rate.

  • Black: Fills the outside of the image with a solid color, black.

  • Replicate: Fills the outside of the image with the last pixel.
Masking Mode Yes

Specifies how a mask is applied to the sampled image. A mask is used to limit areas of the image processed by the tool. Masks can be set after training, however, setting them before training helps the learning phase.

  • Transparent: Collects samples within the unmasked parts of the image, but also collects data from the Context Region from the areas that are masked. This ensures that features or defects that are at the boundary of the mask or ROI The Region of Interest (ROI) defines the area of operation for the tool. The ROI preserves the positions, angles, stretch and skew of the original image. generate the same response as features or defects that are in the center.
  • Mask: Ignores the pixels in the masked parts of the image during training and runtime. This setting also alters the response of the tool to defects or features near the mask boundary. This setting also works to focus the tool toward the center of the ROI.
  • Overlay: Adds the mask to the sampled image as an additional color channel.
Centered Yes This parameter can be used when your inspection contains objects that have a circular symmetry in respect to the center of the view A view of an image is a region of pixels in an image. Tool processing is limited to the pixels within the view. You can manually specify a view, or you can use the results of an upstream tool to generate a view., such that the resulting learned model becomes simpler because the object can be virtually unrolled.

Training Parameters

The Training tool parameters control the training process.

Note: If you change the training parameters of an already trained tool, you must retrain the tool.
Parameter Expert Mode Description

Training Set

 No

Specifies the training data set used to create a neural network model. The system only extracts the features of the images that are included in the training set.

Click Edit to open a dialog where you can specify the sample sets and the percentage of the labeled images used as training samples. The system selects the training images randomly each time you start training.

Epoch Count

 No

Specifies the number of optimization iterations done during training. Use a lower value when your application has limited complexity, or when creating a lower quality model for testing different parameter settings.

The tools typically need to see the training image set A collection of images of your specific application. A training image set represents images of a specific part or process acquired in a consistent way using the lighting, optical, and mechanical characteristics of your runtime system. The training image set includes images that represent the range of image appearances that you expect to see in normal operation. data about 50 times, which is the default setting, and is typically sufficient for most standard applications.

If you use fewer epochs, the neural network might get stuck in learning, or become unable to accurately solve the problem. However, if you use too many epochs, you risk overfitting Overfitting means that the tool is memorizing the data from the training images, which leads to lower performance when using unseen images..

Note:
  • If your training image set contains a large number of images, or the size of the images in your training image set are large compared to the Feature Size, you might need to increase the Epoch Count setting.
  • There is a risk of overfitting if you have a very small training image set. If you have a large training image set, and your network is addressing all of the samples in the set, the network is much more general than if you have a smaller training image set and the network is keyed into that small sample set.
Capacity No

Specifies the capacity of the model to account for different degrees of visual complexity. For very simple patterns, use a lower Capacity setting to prevent overfitting. For complex images or objects, use a higher setting to prevent underfitting.
Training Passes No

Specifies the number of iterative trainings to perform in order to better cope with statistically difficult problems which show varying complexity across images, and/or within different areas of the images. This parameter helps to guide the tool to more complex areas in the images.

Complex Areas =

Simple Areas =

With a higher Training Passes setting, the tool focuses on the complex areas in the image and ignores simple areas. Training time increases, but the processing time remains the same.
Low Precision No Specifies whether or not the tool creates a low precision model after training. The low precision model is useful for applications where speed optimization is a priority during runtime operation. The results might be different between the two modes.

When enabled, a white lightning bolt icon appears in the icon of the tool.

Note:

The low precision model requires the Standard or Advanced license, and also requires a GPU with CUDA Compute Capability 6.1 or higher (for more information, visit the NVIDIA website).

Perturbation Parameters

With the Perturbation parameters, you can allow the tools to generate images that simulate the variations expected during runtime.

For example, if your part rotates by +/-45 degrees, you can set the Rotation parameter, and the software rotates the images during training by that amount. However, image perturbation Perturbation is the process of improving the trained tool tolerance of part and image variation by simulating the effect of specific types of variation. is not a substitute for collecting and training actual images. The image perturbations can only approximate the variations of real parts or scenes.

Note: For the best results, Cognex recommends using actual sample images of the part variations. When a part spins, there might be different shadows, which are not captured by the tool when the rotation is based on artificially spinning the image.

The Perturbation parameters are common across all of the tools, and you can use these parameters in combination. This way, you can generate more complex images by using the parameters separately, as well as in conjunction.

Note: All training image samples can be applied to perturbation. Every train image has 50% chance to be augmented to every checked options independently. By the law of large numbers, all perturbation options can be applied equally and trained by increasing the number of epochs.
Parameter Expert Mode Description

Rotation

 No

Specifies the possible orientations, defining a piece-wise uniform distribution from which a perturbing rotation angle is drawn (if set to [0°, 0°], no rotational perturbation is used).

For example, with the following settings, [-10°, 10°] or [170°, 190°], the part can be detected between -10 and 10 degrees, or between 170 and 190 degrees. These types of settings are particularly useful if your part can be orientated with a regular orientation, or upside down with a small variation in its angle, but never in between.

Scale

 No

Specifies the normal distribution around 1 from which a perturbing scaling factor is drawn (if set to 0%, it is not applied).

For example, a Scale setting of 100%, provides images half the original size to double the original (0.5x to 2.0x the original).

Original = Possible Perturbations =

Aspect-Ratio

 No

Specifies the normal distribution around 1 from which a perturbing aspect-ratio factor is drawn (if set to 0%, it is not applied).

Shear

 No

Specifies the normal distribution around 1 from which a perturbing shearing factor is drawn (if set to 0%, it is not applied).

Flip

 No

Specifies whether to randomly flip the image during sampling in the horizontal, vertical or both directions. This is useful if the object or surface exhibits the corresponding symmetries.

This setting can help to highly increase the quantity of training data.

Horizontal
Vertical
Both

Luminance

 No

Specifies the normal distribution around 1 from which a perturbing luminance factor is drawn (if set to 0%, it is not applied).

Contrast

 No

Specifies the normal distribution around 1 from which a perturbing contrast factor is drawn (if set to 0%, it is not applied).

Invert Contrast

 Yes Randomly inverts the contrast in order to simulate contrast reversal in the training samples.

Processing Parameters

The Processing parameters control the way the tool processes images. Processing with the same models always gives the same results. You do not need to retrain the tool after changing the Processing parameters. You can see the effect after reprocessing the database.

Parameter Expert Mode Description

Sampling Density

 No

Specifies the density of sampling points with respect to the Feature Size setting. The sampling density The amount by which the sampling point is moved in the image between samples, expressed in terms of the feature size. If the feature size is 100 and the sampling density is 1, the sampling location is advanced by 100 pixels between samples. A sampling density of 4 advances the sampling location by 25 pixels between samples. determines the degree of overlap between adjacent samples. A sampling rate of 1 means that the sampling location is incremented by the feature size between samples. The default sampling rate for most tools is 3, which means that the sampling location is incremented by one third of the feature size.

The Sampling Density parameter affects the processing time and the accuracy of the tool. Higher Sampling Density settings are more accurate but significantly increase processing time. The processing time depends approximately quadratically to the Sampling Density (Sd) value. For example, a Sampling Density setting of 1 is approximately nine times faster than a setting of 3 (n2).

Iterations Yes

Specifies the number of times the tool will search locally to find stronger anomalies. In contrast to the Sampling Density parameter, processing time only depends linearly on the number of iterations.

Threshold

 No

Specifies the threshold that determines whether the tool detects and marks regions as good or bad. The tool classifies values below threshold as good, and above the threshold as bad. You can also set the threshold value using the Score graphic in the Database Overview.
Simple Regions No Specifies that the tool can only extract "simple regions", in other words, polygons without holes.
Region Filter Yes Specifies a filter for the tool to use as criteria for found regions. This filters regions from the results that do not match the criteria. If you leave the parameter empty, the tool returns all regions. To apply a region filter, reprocess the tool after setting the filter phrase.
Note: The syntax for filters is the same as the syntax used for Display Filters. For more information about constructing the syntax for a filter, see Custom Display Filters.

 

VariablesVariable TypeDescription

score

Floating Point

Returns the highest defect probability among the probabilities of the pixels in the region.

holes

 

area

Returns the number of pixels in the defect region.

Example: area>= 150

perimeter

 

compactness

Returns the equivalent of: 4 * π * area / perimeter^2.

coverage

Returns the overlapping ratio between labeled area and marked area.

Example: coverage>= 0.5

x

Returns the x-axis position of the center of the defect region.

Example: x>150

y

Returns the y-axis position of the center of the defect region.

Example: y>50

name

String

 

You can apply region filters to the result of the confusion matrix depending on the setting of the Count dropdown in the Database Overview.

For more information, see Count Options and the Confusion Matrix.
Use Baseline  

Specifies whether or not the tool uses a weight on the scoring on the different regions of the image, depending on the variation over this region, measured in all images. This parameter lowers the response with the average noise level. This is useful when the view contains multiple distinctive structured areas.

High Noise Areas
Low Noise Areas

This allows us to use a single Threshold for both areas.