Release History

This topic lists the features and resolved issues in previous releases of VisionPro Deep Learning.

VisionPro Deep Learning 1.1.0

Red Analyze High Detail Mode

A new solution with a powerful new architecture for supervised segmentation tasks that were previously only handled by Red Analyze Supervised Mode. With the addition of Red Analyze High Detail Mode, Red Analyze tools are now reorganized and fall into the following category:

Name	Architecture	Mode
Red Analyze High Detail	High Detail	Supervised
Red Analyze Supervised	Focused	Supervised
Red Analyze Unsupervised	Focused	Unsupervised

Loss Inspector

A monitoring tool of the validation loss change during training that guides you in training High Detail Modes (Red Analyze High Detail and Green Classify High Detail mode).

F1 Score of Confusion Matrix for Red Analyze tools
- The F1 score of the confusion matrix is introduced to show the result summary of the confusion matrix in Red Analyze tools.
- Unlike the precision, recall, and F-score in Region Area Matrix which are calculated on a "pixel" basis, the F1 score of a confusion matrix is calculated from each "item in the confusion matrix table", aiming to suggest the performance of this tool with respect to the confusion matrix result.
Auto Threshold for Red Analyze High Detail and Red Analyze Supervised
- The T1 and T2 threshold values that maximize the F1 score of confusion matrix on Database Overview by following one of the 4 criteria in the Count dropdown menu. These threshold values are automatically calculated for you by enabling a checkbox.
Region List for Red Analyze tools
- The region list is introduced to help you find and check the defect regions (both labeled defects and markings) in a view more easily and conveniently and thus enhance your usability of Red Analyze tools.
Overlay Checkbox
- For all tools, the overlay checkbox is introduced for a handy alternative to "Cycle through Overlays" in the right-click pop-up menu on Image Display Area. See an example of the overlay checkbox for Blue Locate tool.
Restore Parameters
- For all tools, the restore parameter button is introduced for an easier "roll-back" of the set of parameter values in the Tool Parameter pane toward those that were used in the last training session. See an example of the restore parameters button for Blue Locate tool.
Wibu CodeMeter Runtime update from 6.70a to 7.20b to resolve the known security vulnerabilities (WIBU-200521-01, WIBU-200521-02, WIBU-200521-03, WIBU-200521-05, WIBU-200521-06).

Resolved Issues 1.1.0

Multiple GPUs are now fully supported efficiently for High Detail modes. See Multiple GPU Utilization for the details.
Fixed a bug that the multiple additions of the same plugin with Manage Plugins.
Fixed a bug that VisionPro Deep Learning is not launched when the version of nvml.dll (C:\Windows\System32\nvml.dll) is lower than the NVIDIA graphics driver version.
Fixed a bug in Blue Read that the overlay of marking and label is not correctly displayed.
Fixed a bug in Red Analyze that the area under ROC curve is different from AUC value in Database Overview - ROC tab.
Fixed an issue in Designer for Red Analyze that the workflow of Deep Learning Runtime Block of Red Analyze slows down due to heatmap processing. It is fixed by adding a 'heatmap' checkbox and the users now can control the processing speed by enabling/disabling the checkbox.
Fixed some naming inconsistencies in C:\ProgramData\Cognex\VisionPro Deep Learning\1.1\Examples\c#
- The assembly name for Example.Training.HDRed.Console is modified to the right name.

The project file name "Example.ViDi.Suite.GUI.Plugin" is changed to "Example.Deep.Learning.Studio.GUI.Plugin"

The assembly name with space delimiter is changed to the one with "." delimiter.

VisionPro Deep Learning 1.0.1

Support for NVIDIA GeForce® RTX™ 30 series.
- VisionPro Deep Learning 1.0.1 supports RTX 3090, RTX 3080 and RTX 3070 for all tasks.
- If you're exploiting multiple GPUs, all of them should be consist of RTX™ 30 series of the same type and of the same specification. See Use of Multiple GPUs for more details.
- Currently, there are some known issues relating to NVIDIA driver versions and Processing Time. See Known Issues for more details.
Support for Optimized GPU Memory in Parameter Search

If you use Focused Mode in Parameter Search, you can set and use Optimized GPU Memory.
If you use High Detail Mode in Parameter Search, Optimized GPU Memory is automatically disabled for the efficiency of High Detail Mode.

Resolved Issues 1.0.1

Resolved an issue that the prediction result of a cloned Red Analyze Tool is incorrect when Low Precision is enabled.
Resolved an issue that labeling is mistakenly applied on images, not views, after receiving views from an upstream tool.
Resolved an issue that the space between "Image Record" blocks you from using ImageRecord feature with DeepLearningRuntimeBlock inside Designer.

VisionPro Deep Learning 1.0.0

Two modes for the Green Classify Tool.

High Detail mode (from SuaKIT), which provides higher classification accuracy on many challenging applications.
Focused mode (from ViDi) for simpler applications that need faster training or runtime execution.
A Heat Map for High Detail mode, which visually indicates which portions of the image were most influential in the classification decision.
Validation sets for High Detail mode. These provide feedback on model performance and are used for training optimization.
You can enable/disable GPU memory optimization for use with High Detail mode and other tools.

Simpler labeling controls, including

Selecting multiple samples when label or accepting views, clearing markings and labels, processing, or managing image sets.
Automatically re-using the last label name by default, rather than requiring you to explicitly type each new label.

Resolved Issues 1.0.0

Resolved an issue with feature size indicator does not change after changing the value of the feature size in the sampling tool parameter.
Resolved an issue with "Save new version" in the workspace list does not act in the Blue Read Tool.
Resolved an issue with name of Set are shown as original name in the image viewer even though users change the name.

ViDi 4.1.0

Additional Blue Locate tool enhancements, including the following:
- New handles on the feature graphic labels to make resizing easier.
- Select either rectangular or circular feature graphic labels.
- Display or hide the name of feature graphic labels.
- Copy and paste feature label graphics, as well as cloning a feature graphic label by clicking on it.
- Auto-generate the creation of a layout model by selecting features and selecting Create Model.
Additional configuration options for assigning the GPU mode and memory, along with an auto-save feature and training image cache path, can now be set from the Options dialog, which is available in the Starting Cognex ViDi Suite startup dialog.
The ViDi Suite GUI has been localized into six languages: French, German, Japanese, Korean, Simplified Chinese and Spanish. The language can be selected from the Locale drop-down within the Options dialog.

Resolved Issues 4.1.0

To improve the ability to determine if a view is labeled or not, a new property, IView.IsLabeled, has been added to the .NET API.
Resolved an issue with large images and the Red Analyze tool in Unsupervised mode that could result in an invalid exception "Database is Locked" error.

ViDi 3.4.1

Blue Locate tool
The Blue Locate tool has undergone significant overhaul, including the following:

Create a Layout Model, which provides presence/absense verification of features in areas of an image.
New label, marking and model graphics to both increase visibility and usability by making labeled and marked features clearer, and handles to enhance graphic interaction when moving and resizing labels.
Feature identifiers can now be up to 140 characters.
Improved capabilities for oriented and scaled features.
Improved workflow when labeling images/views, with the tool remembering the previously labeled feature, allowing you to start with one image, create a label, and then apply that label to remainder of your images.
Improved performance of node models, along with a new Model Editor GUI.
The tool can now create a region of interest for downstream tools based on a feature, a match, or a node or layout model.

Training workflow

There is a new paradigm for defining the images/views that will be used to train tools, allowing the process to be more transparent and controllable through integration with the new image sample sets.

Image sample sets

The ability to create and manage images/views as training and testing sample sets has been added, which allows for better management of the images/views used to train and test tools.

Database Overview filtering

While in Expert Mode, it is now possible to filter results and statistics in the Database Overview, using the syntax of the Display filters.

Blue Read tool

Capability to identify and read rotated characters.

Red Analyze tool

An eraser tool has been added which allows you to remove areas of defect regions and masks.

In addition to the above new features, there have been numerous performance and stability improvements.

Resolved Issues 3.4.1

Resolved an issue where exporting or cloning a large workspace could cause the ViDi Suite GUI to become unresponsive or unexpectedly quit.
Resolved an issue where a Blue Locate or Blue Read tool's graphical display of markings could be offset from its actual location due to the language settings of the user's PC.
Resolved an issue with the Blue Read tool's Label Views dialog not displaying the names of Regex models that had been created.
Resolved an issue with the Red Analyze tool's generated reports not displaying the markings in the report.

ViDi 3.3.0

General stability and performance improvements.
Improved support for NVIDIA® Turing™ Tensor Cores for the Red Analyze and Green Classify tools with the Low Precision parameter enabled.
Two new model types have been added for the Blue Read tool.

A regular expression model, which allows you to specify string matches based on regular expressions. This type of model offers faster, more flexible model matching for applications with linear arrangements of text than the existing node model.
A string model, which allows you to match linear strings based on the length of the text string.

Improved performance for image displays that include a large number (500+) of feature graphics.
Improved readability of the Green Classify tool's class labels when creating reports for the tool.
Simplified controls for rotating a tool's Region of Interest (ROI).
Support for Cognex VisionPro version 9.5 has been added.

Resolved Issues 3.3.0

Issue	Description
AIC-1861	Corrected an issue with NVIDIA drivers 411.31 and newer. For more information, please visit the Cognex Support Site VisionPro ViDi Software Driver Support page.
AIC-1801	The Set Group option has been restored to the Actions for N Views menu in the View List, which allows you to create a group of images when using the Red Analyze tool in Unsupervised Mode.
AIC-1201	Corrected an issue where a workspace would not function correctly if a stream contained a single quote in the name.
AIC-1223	Corrected an issue where the ViDi Suite GUI would become unresponsive when applying a region for a Red Analyze tool with numerous views.
AIC-1387	Corrected an issue where the ViDi Suite GUI would not terminate properly after attempting to connect to a runtime-only service.

ViDi 3.2.1

This release introduces the ViDi Client/Server functionality, allows you to use the training and runtime implementations of the ViDi software on multiple PCs, through the use of a server/client model. This functionality gives you the option of using a remote service for runtime inference and/or training, instead of using the GPU on your local PC. This can be useful if you have a more powerful GPU on the server, or perhaps you do not have a GPU installed locally.
This functionality supports up to 4 clients for deployment operation, and up to 8 clients for development.
This release provides enhancements and speed improvements for the ViDi runtime:

Support for NVIDIA® Tensor Cores has been added, and the GPU Optimization feature also now supports Tensor Cores.
The Low Precision parameter has been added to the Red Analyze and Green Classify tools, which optimizes the tools for speed during runtime operation.

The Feature Filter has been added to the Blue Locate and Blue Read tools, which allows you to filter found features by ID, X, Y, size and orientation.
The ViDi GUI and the ViDi Help file have been translated into Simplified Chinese. To view the GUI and Help in Chinese, the OS language setting must be set to Simplified Chinese.
Support for up to 32MP images (up to 16k in height or width) has been added.

ViDi 3.1.0

Tool and GUI Changes

The ViDi Blue Read Tool has been added for Optical Character Recognition (OCR) applications.
Speed improvements and increased performance for the Blue Locate Tool when matching models.
The Feature Size parameter is now interactive and can be graphically resized within the GUI.
Labeling with the Blue tools has been improved to include multi-feature selection and interaction.
The Blue tools have improved flexibility for model optimization.
In the results, a Confusion Matrix has been added for the Blue Locate and Blue Read tools.
A new feature/match and region-focus display mode has been added when using the Database Explorer.
Checks have been added to provide feedback on wrongly labeled views.
The examples have been updated, and there is a new example for the Blue Read tool.
The Green Classify Tool now includes the Exclusive parameter, which, when disabled, allows you to label a single image with multiple labels (indicating that the image belongs to multiple classes).
The Red Analyze Tool – Statistics have been enhanced for better readability.
There have been numerous issues resolved, and additional enhancements and improvements to the ViDi software.

Processing and Support Changes

The GPU Memory Optimization is now activated by default (when using either the Windows Display Driver Model or Tesla Compute Cluster drivers). This feature can be activated and/or deactivated via command line arguments and the API.
Command line arguments have been added to allow you to select the number of GPUs to utilize at startup of ViDi, to specify the GPU mode and to specify the amount of GPU memory to pre-allocate.
There is a new Cognex VisionPro wrapper assembly, which provides data conversion between VisionPro and ViDi Suite, offering the following:
- Efficient image conversion
- Combined graphics
- Fixturing conversion
- Calibration conversion
Support for Cognex Designer 4.0 has been added, with the addition of a ViDi Sub Task block, which contains the ViDi Suite tools and the ability to program the tools within the Cognex Designer IDE.
Support for Cognex VisionPro 9.3 has been added.
Support for Microsoft® Windows® 10 Professional (64-bit).
Support for Microsoft® Windows® 7 Professional (64-bit).
The performance when loading training Workspaces, utilizing on-demand loading of tools, has been improved.
A new Workspace archive format, based on ZIP files, has been implemented.