Cognex Deep Learning Release Notes

Release Notes
Release Notes
VisionPro Deep Learning 3.2.1 is a hotfix release of VisionPro Deep Learning 3.2
Enhancements
l The AI Labeling Tool can now be used in the service client environment.
l The AI Labeling Tool now has the undo function.
l The Green Classify High Detail and the Green Classify High Detail Quick tools now also support processing with
images smaller than 64 x 64 pixels.
l The Green Classify High Detail and the Green Classify High Detail Quick tools now also support training only with
images smaller than 64 x 64 pixels.
Note:
l The VisionPro, Designer and VisionPro Deep Learning software require that a valid Cognex Security Dongle be
installed directly to PCs running the software during all phases of operation (programming, processing, training,
testing, etc.). Any attempts to temporarily remove, substitute, or share a Cognex Security Dongle may cause
your system to operate incorrectly, and may result in the loss of data.
When VisionPro Deep Learning is configured for the Client/Server functionality, and a computer has been
configured as a server, the Cognex Security Dongle must be attached to the server, but the client(s) does not
need a Cognex Security Dongle.
l Sample code is included with the VisionPro Deep Learning 3.2.1 installer. The resources for the sample code
(VisionPro Deep Learning 3.2.1 Sample Code Resource.msi), which include images and workspaces, are
available for download from the Cognex Support Site.
l For more information about new API functionality in this release, see the Cognex VisionPro Deep Learning Suite
.NET Library and Cognex VisionPro Deep Learning Suite C API help files.
1
Resolved Issues
Resolved Issues
l Fixed an issue where the 100 x 100 ROI value set by the user was randomized in the ROI setting window.
l Fixed an issue where the VisionPro Deep Learning software was slow when loading the database overview when the
number of classes was eight or more.
l Fixed an issue where the resize size set by the user could not be imported into the Parameter Search in Auto mode.
Previously, the manually set resize size was overwritten by the size calculated by Parameter Search, but now this
does not happen anymore.
l Fixed an issue with the Green Classify High Detail and the Green Classify High Detail Quick tools where the score
was different between the training API and the runtime API. Now both show the same score.
l Fixed an issue with the Green Classify High Detail Quick tool where the marking did not change to label after clearing
a label in a view which had marking and accepting the view.
l Fixed a compatibility issue between VisionPro Deep Learning 3.2.0 and VisionPro 9.20
l Fixed an issue where the Green Classify High Detail mode parameter Validation Set Ratio could not be changed from
20%.
2
Known Issues
Known Issues
The following issues have been identified in this release. For the API known issues, see API Known Issues on page 47
VisionPro Deep Learning 3.2.1

There is a score difference between the GUI and the runtime when using 16-bit images with the Green Classify High Detail,
Green Classify High Detail Quick, and the Red Analyze High Detail tools.
VisionPro Deep Learning 3.2
GPU
l Processing Time could be slower compared to 1.0.0 under old GPUs. Processing Time could be slower compared to
1.0.0 when using NVIDIA GeForce® RTX™ 20 series or older series due to NVIDIA internal issues.
l Due to several issues related to GUI and training, all users of VisionPro Deep Learning 3.2.1 MUST use NVIDIA
driver that is of:
o After VisionPro Deep Learning 3.2.1, recommended GPU driver version range is from 528.02 to 532.03, with
the tested version being 528.02. Using driver versions beyond 532.03 may result in slower training and
processing times in High Detail .
Cognex Deep Learning Studio (GUI)

l During creating a report when you log off your PC, some of images are shown black. To avoid this issue, please do
not log off your PC (neither turn off the remote environment nor lock screen) while you are creating a report.
l “No License Found” error occasionally occurs when you try to open 2 different versions of VisionPro Deep Learning,
or launch Parameter Search and VisionPro Deep Learning, at the same time.
Workspace
l Boost::filesystem::rename error occurs when the user or system repetitively saves the workspace in a short time.
This is just an alert that there was an issue while saving. You can solve this problem only by saving the workspace
again.
l Group is shown in the workspace imported from ViDi Suite.
o If you import workspaces including "Group" feature which are made 4.1 and before, "Group" is still displayed
in Cognex Deep Learning Studio 3.0, but you cannot modify this.
o See the FAQ for the workaround for this issue.
o It is recommended to delete "Group" in the previous workspaces and import to Cognex Deep Learning Studio
3.0.
l Especially with bigger than 10GB workspaces,
o Cloning work does not end when you clone tools in different big workspaces at the same time.
o Some images are not included in the created report.
3
Known Issues
Tools - General
l If a tool is not saved well because of unspecific reasons, you cannot export or clone the workspace or the tool. You
can avoid this problem by processing the tools again and save the workspace.
l For all tools, if the tool crashes with some unknown reason, images in this tool are deleted when the workspace that
contains this tool is imported.
l If the name of a set in Database Overview contains a single quote ('), you cannot change the name of this set.
l If the name of a set in Database Overview contains a single quote ('), an error occurs when you try to filter views on
View Browser by clicking the icon of this set in Database Overview.
Tools - Red Analyze

l Processing results are different between GPU and CPU. The difference is in the second decimal place of the result.
For example, When you process with CPU and GPU respectively, different results can be obtained, such as 0.80 and
0.81.
l Defect regions do not extend to the left side of the view in Red Analyze Tool. To avoid this problem, when the defect
area touches the left side of the image, please set 4 pixels wider ROI than the original image size. For example, if you
have (x=0,y=0,w=100,h=100) image and it has detected area touch to left side of the image, please set (x=-4, y=0,
w=104, h=100) ROI size.
l When you set the smaller feature size (especially smaller than 10) and train the Red Analyze tool (both in Supervised
and Unsupervised mode), the workspace is getting bigger. The vvb file which is for saving heat map within the
workspace is the main reason. For example:
o feature size=10, sampling density=3, 8192x8192 image: 1 vvb file size=94MB
o feature size=10, sampling density=10, 8192x8192 image: 1 vvb file size=1GB
o feature size=40, sampling density=3, 8192x8192 image: 1 vvb file size=23KB
l For Red Analyze Supervised modes (High Detail and Focused), applying the region filter could affect the result of
Auto Threshold to be compromised in generating an optimal threshold for the best F1 score.
l For Red Analyze High Detail, changing a value in Processing Parameter in Tool Parameters and rolling back to the
original value could affect the result at 4 decimal places.
l For Red Analyze High Detail, generally, you can guess that processing time increases when you reduce the
Downsampling Size, but the opposite case can happen but this is not a bug.
l For Red Analyze Supervised, if there were an extremely enormous amount of defect regions per view, especially
when the tool is not trained enough, VisionPro Deep Learning becomes slow and the expected value of processing
time becomes inaccurate. Please take the tool re-trained to learn the image information fairly enough.
l For Red Analyze Tool, the area of a marked defect region and the area of a labeled defect region which is created by
accepting (right-click on a marked region → click Accept Region) that marked region, is slightly different from each
other.
l For Red Analyze Focused Supervised, its Database Overview crashes when there are multiple types of labeled
defect regions.
4
Known Issues
Tools - Green Classify

l Once you are done with training High Detail mode of Green Classify Tool, you cannot change the name of the class.
l For Green Classify High Detail, resizing views takes a little more time when it is the first processing of the tool.
l For Green Classify High Detail, the processing preparation takes a little more time when it is the first processing of the
tool.
Tools - Blue Read

l For Blue Read, the markings are displayed on the view regardless of the Marking checkbox in Image File Information
is whether enabled or disabled. This issue has existed since ViDi 4.1.
Parameter Search
l In Home page, if you set 1-depth directory path (example: "E:/," or "A:/") in the Path, Parameter Search is terminated
with an error and you can't re-open it. Please refer to Unable to start Parameter Search after setting 1-depth directory
path at Path in its Home page for the workaround of this issue.
VisionPro Deep Learning Service

l MultipleDevicesPerTool is no longer supported for VisionPro Deep Learning. Therefore you will have an “One or
more error occurred” error if you select “--gpu-mode =MultipleDevicesPerTool” for service. Please choose another
option to run VisionPro Deep Learning Service.
l For VisionPro Deep Learning Service, Export Runtime Workspace to Service on the right-click pop-up menu on
Workspaces panel is currently disabled. To export runtime workspace, please go to and select Workspace - Export
Runtime Workspace to Service instead.
l For VisionPro Deep Learning Service, when hosting a service with 2 different server ports (Client 1 connects via Port
1 and Client 2 connects via Port 2), the “Lock” icon does not appear on a workspace in Workspaces for a client when
this workspace has been already used by one another.
VisionPro Deep Learning 2.1 (2.0, 2.1.1)
GPU
5
Known Issues
Operating System (OS)

l Due to OS compatibility issues caused from NVIDIA drivers, Windows Server 2016 is not supported on VisionPro
Deep Learning. For the users with Windows Server 2016 environment, upgrading from Windows Server 2016 to
Windows Server 2019 is required for the utilization of VisionPro Deep Learning servers. Windows Server 2019 is only
supported for 'servers' configured to use the Deep Learning Client/Server functionality.

l “No License Found” error occasionally occurs when you try to open 2 different versions of VisionPro Deep Learning,
or launch Parameter Search and VisionPro Deep Learning, at the same time.
Workspace
again.
in Cognex Deep Learning Studio, but you cannot modify this.
o It is recommended to delete "Group" in the previous workspaces and import to Cognex Deep Learning Studio.
Tools - General
6
Known Issues
Tools - Red Analyze

0.81.
w=104, h=100) ROI size.
other.
l Erasing maskings/regions in the mask/region editor pane (Edit Mask or Edit Region in right-click popup menu on the
image display area) applies Gaussian smoothing on all the regions/maskings. This causes the entire
maskings/regions to shift slightly and their sizes to be changed slightly. This issue has been an existing issue from
VisionPro ViDi.

Tools - Blue Read

7
Known Issues
Parameter Search
l In Home page, if you set 1-depth directory path (example: "E:/," or "A:/") in the Path, Parameter Search is terminated
with an error and you can't re-open it. Please refer to Unable to start Parameter Search after setting 1-depth directory
path at Path in its Home page for the workaround of this issue.

Migration from SuaKIT or The Past VisionPro Deep Learning Versions

l High Detail mode uses a different resize algorithm to SuaKIT (Green Classify Tool in High Detail mode only). To use
an imported SuaKIT model with VPDL runtime API, please use SuaKIT resize API before processing.
l For SuaKIT plugin, The result of processing between the imported SuaKIT Segmentation project in VisionPro Deep
Learning and the original SuaKIT Segmentation project could be different due to the internal logic change of
VisionPro Deep Learning 1.1.
Integration with VisionPro

l Due to the currently limited support of VisionPro 10, the external ROI is not supported for a DLRuntimeTool Edit
Control Panel in VisionPro 10 for the time being.
Plug-in
l For Designer, the added heatmap checkbox is initialized on Deep Learning Runtime Blocks of the tools other than
Red Analyze. Tools other than Red Analyze were supposed not to support heatmap function.
8
Known Issues
VisionPro Deep Learning 1.1 (1.1.1)
GPU
l Internal issues on supporting NVIDIA driver versions. Due to internal issues related to NVIDIA drivers, all users of
VisionPro Deep Learning 1.1.0 MUST use NVIDIA driver that is of:
o 461.09 or higher version for NVIDIA GeForce® series.
o R460 U2 (461.09) or higher version for NVIDIA RTX / Quadro® series.
l Due to internal issues related to runtime, C API may not work as intended. To prevent this issue, all users of C API
MUST follow the steps below.
1. Remove the currently installed NVIDIA driver from your PC.
2. Reboot your PC and install the DCH NVIDIA driver of which version is compatible with VisionPro Deep
Learning 1.1.0 (461.09 or higher, or R460 U2 (461.09) or higher for NVIDIA RTX / Quadro®)


Workspace
again.
9
Known Issues

Tools - General
Tools - Red Analyze

0.81.
w=104, h=100) ROI size.
other.
10
Known Issues
VisionPro ViDi.

Tools - Blue Read



l For SuaKIT plugin, The result of processing between the imported SuaKIT Segmentation project in VisionPro Deep
Learning and the original SuaKIT Segmentation project could be different due to the internal logic change of

11
Known Issues
Plug-in
l For Designer, the added heatmap checkbox is initialized on Deep Learning Runtime Blocks of the tools other than
Red Analyze. Tools other than Red Analyze were supposed not to support heatmap function.
VisionPro ViDi 3.1.0 ~ VisionPro Deep Learning 1.0.1
GPU
l Internal issues on supporting NVIDIA driver versions. Due to internal issues related to NVIDIA drivers, all users of
VisionPro Deep Learning 1.0.1 MUST use NVIDIA driver that is of:
o 457.09 or higher version for NVIDIA GeForce® series.
o R460 U2 (461.09) or higher version for NVIDIA RTX / Quadro® series.
l Due to internal issues related to runtime, C API may not work as intended. To prevent this issue, all users of C API
MUST follow the steps below.
1. Remove the currently installed NVIDIA driver from your PC.
2. Reboot your PC and install the DCH NVIDIA driver of which version is compatible with VisionPro Deep
Learning 1.0.1 (457.09 or higher, or R460 U2 (461.09) or higher for NVIDIA RTX / Quadro®)
l Multiple GPUs are not supported efficiently for High Detail mode. You cannot get an advantage of using multiple
GPUs when you proceed training or processing the other tool specifically with Green Classify Tool in High Detail
mode at the same time. For example, when training or processing job of Focused mode and High Detail mode are
simultaneously being processed, each job could be slower than just training or processing one job at a time.


12
Known Issues
Workspace
again.
Tools - General
Tools - Red Analyze

0.81.
w=104, h=100) ROI size.
other.
VisionPro ViDi.
13
Known Issues

Tools - Blue Read




14
Migrating to This Release

This topic contains information about changes that may affect existing VisionPro Deep Learning projects when migrating
projects from previous releases to the current release.
3.0 → 3.1
Re-optimization Required When Using TensorRT

When using TensorRT with VisionPro Deep Learning 3.1, please note that even if you have already optimized TensorRT with
VisionPro Deep Learning3.0, you must optimize it again for VisionPro Deep Learning3.1.
2.1 (2.1.1) → 3.0
Removal of "Extra Large" from Green Classify High Detail Network

Models
Due to the Green Classify High Detail architecture update in VisionPro Deep Learning 3.0, the Extra Large option in Network
Model parameters for a Green Classify High Detail is removed from VisionPro Deep Learning. If you want to use a Green
Classify High Detail in VisionPro Deep Learning 3.0, saved with the Extra Large network size in the previous VisionPro Deep
Learning versions, import the tool and change its Network Model parameter from a blank to another one (Small, Normal, or
Large).
Removal of Red Analyze High Detail Quick

Due to internal reasons, Red Analyze High Detail Quick is not supported in VisionPro Deep Learning 3.2.1. When you open a
Red Analyze High Detail Quick tool trained in an earlier VisionPro Deep Learning version on VisionPro Deep Learning 3.2.1:
l This tool will be opened as a Red Analyze High Detail tool.
l The trained network of this tool will be lost.
l The views and their labels of this tool will be preserved.
l The Tool Parameter values will be reset to the default values of a Red Analyze High Detail tool.
Also, importing a Red Analyze High Detail Quick tool trained in an earlier VisionPro Deep Learning version into a VisionPro
Deep Learning 3.2.1 workspace is disabled.
High Detail Backwards Compatibility

When you open a Red Analyze High Detail or Green Classify High Detail tool trained in an earlier VisionPro Deep Learning
version on VisionPro Deep Learning 3.2.1:
15

l The Tool Parameter values will be preserved.
When you import a Red Analyze High Detail tool (.vrta) or a Green Classify High Detail tool (.vgta) trained in an earlier
VisionPro Deep Learning version into a VisionPro Deep Learning 3.2.1 workspace:
l The Tool Parameter values will be preserved.
Runtime Workspace Backwards Compatibility

The runtime workspace created in the earlier versions of VisionPro Deep Learning 3.0 that contains a Red Analyze High
Detail, Red Analyze High Detail Quick, Green Classify High Detail, or Green Classify High Detail Quick can't be imported in
VisionPro Deep Learning 3.0. In this case, you need to create a new runtime workspace in VisionPro Deep Learning 3.0.
SuaKIT Segmentation and Classification Projects Compatibility

As it becomes unable to import the High Detail tools created and trained in the earlier versions of VisionPro Deep Learning
3.0, it is also not available to import the trained models (neural networks) of existing segmentation or classification projects
from SuaKIT into VisionPro Deep Learning 3.0. See Import a SuaKIT Classification Project and Import a SuaKIT
Segmentation Project for more details.
2.0 → 2.1 (2.1.1)
Termination of CIE1 Deployment Support

The CIE1 deployment target is officially no longer supported by VisionPro Deep Learning 2.1 as VisionPro Deep Learning
does not support smart cameras. The support for CIE1 deployment target has unofficially been terminated since VisionPro
Deep Learning 2.0, but the option was visible in VisionPro Deep Learning GUI in 2.0.
1.1 → 2.0
Termination of Designer Support

The integration with Cognex Designer is no longer supported from VisionPro Deep Learning 3.2.1. For this particular update,
the users who are required to use Designer should use VisionPro Deep Learning 1.1.
16
1.0.1 → 1.1
Change in Validation Loss Calculation

The base of calculating the Validation Loss in High Detail modes (Red Analyze and Green Classify) is changed from 128
iterations to 0.125 epochs. According to this change, the Patience parameter, which was calculated on 128 iterations (N
times of 128 iterations) basis, is no longer used in High Detail modes and a new parameter named Patience Epoch is used
instead. This new parameter is calculated on an epoch basis (N epochs). For the Green Classify High Detail tools already
trained, you have to reconfigure the value of Patience Epoch manually due to this base change.
Loss Tracking in Log Messages

The records of the validation loss in High Detail modes (Green Classify and Red Analyze High Detail) are no longer available
in Log Messages as its function of providing the validation loss history is replaced with the Loss Inspector.
Region Filter for Confusion Matrix

For all Red Analyze tools, the region filter is now also applied to the result of the confusion matrix. This application of the
region filter is either performed by using filtered() expression.
1.0 → 1.0.1
The workspaces created in VisionPro Deep Learning 1.0.0 are fully compatible with VisionPro Deep Learning 1.0.1
VisionPro ViDi 4.1 → VisionPro Deep Learning 1.0
Architecture of Existing Green Classify Tool

When you open workspaces created from previous version (ViDi Suite) and if there is a Green Classify Tool, you would see
as a Green Classify Tool in Focused mode. Existing architecture of Green Classify Tool is called Focused mode and a new
architecture of Green Classify Tool is called High Detail mode.
Keeping Existing names of APIs

Since VisionPro Deep Learning inherits ViDi, it retains the name ViDi in many parts of the APIs
Internal version of APIs

Product version is 1.0.0 but, 5.0.0 is used for the version in APIs and binaries.
17
No longer supported or Deprecated APIs

GPUMode.MultipleDevicesPerTool is no longer supported. For more information, see API Deprecated Features on page 49 .
Changes to Designer functionality

A project built with the 4.1 (or previous) version of the Designer Deep Learning Plugin is not compatible with the Cognex
Deep Learning Plugin. To force the correct plugin to be used, use the Project Plugin Configurator in Designer’s Utilities.
You cannot train in Designer, you can import Cognex Deep Learning Studio Runtime Workspace(.vrws file) created from
Cognex Deep Learning Studio 1.0 and process the images.
The Runtime Block includes only one output pin. Individual Tools in the Tool Chain are not visible in the UI. And there are no
preconfigured result pins so you need script for every single application.
Deep Learning Runtime Block does not support an External Fixture.
3.4.1 → 4.1
Changes to the Feature Filter Syntax

In the 4.1.0 release, the syntax for the Feature Filter Processing tool parameter was modified. If you were using width and
height in your expression, they must now be qualified with view, such as view.width and view.height.
For example, in version 3.2 and earlier, syntax used to limit the found features to those where the center of the feature lies
within the ROI would have been the following:
x < width and x > 0 and y > 0 and y < height
In 4.1.0 and later, it must now be expressed as the following:
x < view.width and x > 0 and y > 0 and y < view.height
Note:
This syntax can also more easily be expressed by using the in operator. For example, the following syntax could be
used to simplify the expression:
x in [0,view.width] and y in [0,view.height]
Changes to the Viewing of Overlay Graphics

In the 3.4.1 releases, the keyboard shortcuts used to show and hide overlay graphics in the Image Display Area have been
changed. The new keystrokes are the following:
l The Alt key plus the Right Arrow key or the Alt key plus the Left Arrow key
These are used to cycle through the available combinations of display graphics (labels, markings, labels and markings, or no
overlay graphics).
In previous releases, the space bar could be used to show or hide all overlay graphics, but that is no longer supported.
18
Changes to Tool Training Workflow

In the 3.4.0 release, there is a new method for specifying the images/views that will be used to train a tool. If you will be re-
training a tool that was trained in a previous release, you will need to configure the tool to use the new training set paradigm,
via the Select Training Set dialog. Please refer to the Train Neural Network section of each tool.
For the Blue Locate Tool, a change was made in the behavior of X and Y position clamping when using model matching. In
version 2.1, these constraints were interpreted to apply to the position of the origin of the model coordinates. In version 3.0,
they are interpreted as constraints on the position of the centroid of the nodes in the model.
19
Release History
Release History
This topic lists the features and resolved issues in previous releases of VisionPro Deep Learning.
New Features
l AI labeling tool
In the AI labeling tool, you can use two key tools: AI Click Tool and AI Box Tool. These tools simplify labeling by
automatically recognizing and labeling regions, eliminating the requirement for manual drawing. You can find more
details in the "Create Label" section of Red Analyze.
l Red Analyze processing speed
The processing speed of the Red Analyze was enhanced, particularly in Red Analyze High Detail.
l Faster image addition
Adding images takes reduced time, resulting in faster handling when adding a large number of images.
Enhancements
l GPU Clock Stabilizer
The feature to maintain a stable GPU clock in a deployment environment can be helpful in multi-GPU scenarios or in
situations where clock instability leads to reduced processing speed or spikes. For more details, refer to the API
Change Log or API documentation.
Resolved Issues
l Fixed an issue where CPU memory gradually increased when repeating training and processing of High Detail.
l Fixed an issue where an internal error occurred during the creation of a runtime workspace, preventing its creation
when there was a previously trained model using the large network in theGreen Classify High Detail from versions
prior to VisionPro Deep Learning 3.1.1.
l Fixed an issue where High Detail was not supported in ViDi.ITool.RegionOfInterest.
l Fixed an error when training tools with a 1-channel 16-bit image.
l Fixed an internal error that occurred during the Parameter Search when starting the search for Green Classify High
Detail with Large model.

l Hotfix release of VisionPro Deep Learning3.1.
Resolved Issues
l Fixed an issue where tools trained using Green Classify High Detail tool's "large" modes inVisionPro Deep Learning
3.0 were not processing in VisionPro Deep Learning 3.1.1.
l With the use of CLIP, you can now import a JSON file containing a "bad" tag without regions into Red Analyze.
20
Release History
l Using CLIP, fixed an issue where the region name of a "bad" label in the Red Analyze Focused Supervised appeared
as "bad" instead of "defect" during label export and import.
l Using CLIP, fixed an issue of JSON file format labels not being imported in theGreen Classify.
l The tooltip for the "mode" parameter in the High Detail parameters has been updated.

l Windows 11 is now supported
o Now VisionPro Deep Learning 3.2.1 supports Windows 11
l RTX™ 40 series are tested
o VisionPro Deep Learning supports on RTX™ 40 series which are RTX 4070 Ti, RTX 4080 and RTX 4090. It is
tested on RTX 4080 and 4090.
l Minor improvements
o Improved the processing speed for High Detail tool by optimizing the internal logic, including the mask
functionality.
o Due to internal logic improvements, the training speed of Red Analyze High Detail tool has been partially
enhanced.
o In the Auto mode of Parameter search, remaining and elapsed time are displayed during the search.
Resolved Issues
l Fixed an issue in Red Analyze High Detail where the view score would display as 0 after applying a mask and
processing with the API.
l Fixed an issue in the Red Analyze High Detail where the processing results differed when using Whole Image
processing, although differences may still occur within an acceptable margin of error.
l Fixed an issue where out of memory pop-up occurred during processing after completing training with Whole Image
Processing on in Red Analyze High Detail, which prevented saving of the model. The tool now automatically turns off
Whole Image Processing and proceeds with processing if it cannot perform the function.
l Fixed an issue where the application would intermittently crash when loading a runtime workspace with multiple High
Detail and using multi-GPU.
l Fixed an issue where the Reuse mask option was not applied in the ROI settings for downstream tools when Red
Analyze was the preceding tool.
l Fixed an issue where the label check did not run when using Green Classify with more than three tags.
l Restored the pan function that was missing in the image viewer.
l Fixed an issue where Cognex Deep Learning Studio abnormally terminates when clicking on an empty column in the
image list.
l Fixed an issue where Cognex Deep Learning Studio abnormally terminates when entering an invalid color value for
the tag after setting the tag in Green Classify.
l Updated the tooltip description regarding the processing time displayed on the Cognex Deep Learning Studio, as it
was previously shown in view unit instead of image unit.
21
Release History
l Fixed an issue where entering -1 into the command line failed to correctly deactivate the auto-save feature in the
service-client
l Fixed an issue where the C++ sample code, Example.AdaptationMode.HD, was not functioning properly.
l Fixed an issue that caused an error when attempting to switch from Auto mode to Manual mode in Parameter search,
if there were spaces in the tags.
l Fixed an issue where results were not displayed after conducting a search in auto mode with a tool in the downstream
in Parameter search.
VisionPro Deep Learning 3.1.0 (Beta)

l Beta release for Window 11

l Adaptation Mode for High Detail tools
o Adaptation Mode is supported only for Red Analyze High Detail and Green Classify High Detail tool. See
Adaptation Mode: Adapting to Line Variation for details.
l Red Analyze High Detail and Green Classify High Detail Architecture Upgrade
o The neural network architecture of Red Analyze High Detail and Green Classify High Detail is updated in
VisionPro Deep Learning 3.2.1.
l Sampling without "Stride"
o The sampling logic for Red Analyze High Detail training evolved that it can deliver the same performance
without the Stride-based sampling. Due to this upgrade, the Stride in the Training parameters of Red Analyze
High Detail is deprecated from VisionPro Deep Learning 3.0.
l Backwards Compatibility
o For more details, see Migrating to This Release on page 15.
l Change of Validation Loss Calculation
o Another change from this architecture update is the new validation loss calculation for Red Analyze High
Detail, which is used to select the best neural network during validation. See Validation Set and Validation
Loss for more details.
l Automatic Parameter Search
o While the manual mode tests all combinations of the listed parameters and their value ranges, the auto mode
cuts off unnecessary tests based on performance criteria and thus it can save more time for getting you the
best values of provided parameters. The auto mode of is only supported for tools. See Parameter Search for
details.
l Label Check
o The label quality value is scaled from 0 to 100. The lower the value, the noisier or the worse each label is. You
can dramatically reduce the time spent reviewing the labels by sorting images in ascending order by the label
quality value. Label Check is only supported for Red Analyze Focused Supervised, Red Analyze High Detail,
and all Green Classify tool types. See Optimize Labeling with Label Check for more details.
l Cognex Label Import/Export Plugin (CLIP)
22
Release History
o CLIP is a new plug-in that supports importing/exporting masks and labels of a VisionPro Deep Learning
workspace in VisionPro Deep Learning internal and external formats. You can export masks, labels, or
images in a VisionPro Deep Learning workspace in VisionPro Deep Learning internal format or png file
format, and also import masks or labels in various formats into a VisionPro Deep Learning workspace with
this plug-in. See Cognex Label Import/Export Plug-in for details.
l TensorRT Support for Runtime
o The NVIDIA TensorRT is a library that boosts the inference speed of a deep neural network running on
NVIDIA GPUs. The VisionPro Deep Learning now supports TensorRT for the runtime API, boosting the
processing speed of each tool. TensorRT for Runtime is only supported in High Detail tools. For details of
TensorRT support, see API Changelog on page 33.
l Green Classify High Detail Batch Processing Support for Runtime
o Using the Batch Processing API in runtime can speed up the Green Classify High Detail processing via API.
For more details about the batch processing API, see API Changelog on page 33.
l View Browser - List Mode
o The list mode presents an information table that offers much more details about a view's status including
marking, label, sets, or a view's index. You also can star a view if it is an important view or write and edit a
memo for each view. See View Browser - List Mode for details.
l Other Improvements & Changes
o UI/UX Improvements in Parameter Search
o GUI Refinement
o Applying Label Change for Green Classify
o Image Resize for Green Classify High Detail
o Validation Set Indicator and Filter for Validation Set
o Scale Change of Validation Loss
o Image Processing Logic Change for Red Analyze High Detail
o Minimum Image Size Change for Green Classify High Detail
o Training Logic Change with Patience Epochs and Minimum Epochs
o Designer Plug-in Support
o Red Analyze sort filter keyword count(region!) in View Browser which sort the views by the number of
labeled defect regions is changed to count(region![area > 0]).

l Fixed an issue that the execution of VisionPro 10.1 C# API method Cognex.Vision.Startup.Shutdown() is not
completed within an expected time when using VisionPro Deep Learning Tool Block.
l Fixed an issue that the Focused tools (Blue Locate, Blue Read, Green Classify Focused, Red Analyze Focused
Supervised, Red Analyze Focused Unsupervised) occasionally expose inconsistent results given exactly the same
configurations.
23
Release History
l Fixed an issue that the processing of Blue Locate or Blue Read rarely gets slower.
o To solve this issue, we allowed access to a new processing parameter called max_scan_iterations in the
runtime API. With this parameter, you can limit the maximum search time to speed up the processing. For
more details, see API Changelog on page 33.
Resolved Issues
l Fixed an issue that the execution of VisionPro 10.1 C# API method Cognex.Vision.Startup.Shutdown() is not
completed within an expected time when using VisionPro Deep Learning Tool Block.
l Fixed an issue that the Focused tools (Blue Locate, Blue Read, Green Classify Focused, Red Analyze Focused
Supervised, Red Analyze Focused Unsupervised) occasionally expose inconsistent results given exactly the same
configurations.
l Fixed an issue that the processing of Blue Locate or Blue Read rarely gets slower. See API Changelog on page 33 for
more details.

l Red Analyze High Detail Quick and Green Classify High Detail Quick
o Faster than High Detail modes in training
o Stay solid in task accuracy
o More convenient with much less Tool Parameters to be tuned by hand.
l Revamped Parameter Search
o The user interface of Parameter Search has been completely renovated to enhance the usability of the tool.
l Heatmap Overlay on Image Display Area for Green Classify High Detail and Green Classify High Detail Quick
o For Green Classify High Detail and Green Classify High Detail Quick, the heatmaps are not only visible
through View Inspector but directly in Image Display Area by enabling Heatmap checkbox in Image File
Information. See Heatmap and GUI and API for the details.
l Green Classify API Example Codes are updated for getting label information from images (RAPID XML)
o C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c++\Example.Training.HDGreen
o C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c++\Example.Training.Green
l Improved tool loading speed into Workspace for a Red Analyze tool when it has significantly many detected regions
(more than hundreds or thousands of regions) or regions with complex shapes per image.
l Enhanced document contents. See Documentation Updates on page 52 for the details.
Resolved Issues
l Fixed an issue that some runtime API example codes were unable to be run because of missing DLL files.
l Fixed an issue in High Detail Mode tools that the training time keeps increasing when training multiple High Detail
mode tools in sequence.
24
Release History
l Fixed an issue in Parameter Search that searching parameters for a Red Analyze tool already trained in VisionPro
Deep Learning unnecessarily copies existing .vvb files and thus grows the size of the current workspace.
l Fixed an issue in C API that 32-bit images (8-bit, 4 channel) added to a workspace by C API look bizarre when they
are viewed in GUI.
l Fixed an issue in Loss Graph that the Elapsed Time in Status table is not in accordance with the elapsed time in
Progress bar.
l Fixed an issue in Loss Graph that the Elapsed Epoch in Status table and Epoch in the loss graph are displayed
incorrectly.
l Fixed an issue that the value of ViDi2.View.maskCache keeps increasing and consumes memory.
l For High Detail modes, fixed an issue that processing with GPU designation is not available when calling
ISample.Process().
l For VisionPro Deep Learning Client/Server environment, fixed an issue that VisionPro Deep Learning Client/Server
cannot accept "16-bit depth, png/tiff format" image file as its input image.
l For runtime workspaces, fixed an issue that users cannot get the label information of Green Classify High Detail with
Runtime.IGreenTool.KnownTags
l For Red Analyze, fixed an issue that after changing mode of tool operation (Supervised, Unsupervised) the class of a
region is displayed incorrectly.
l For Red Analyze, fixed an issue that after changing mode of tool operation (Supervised, Unsupervised) a region
cannot be modified with Edit Region.
l For Red Analyze High Detail, fixed an issue that if you change the mode of a Red Analyze tool to Red Analyze High
Detail, add images, set ROI, and train with the default tool parameters, the score for a view is not generated.

l VisionPro 10.0 Integration
o VisionPro Deep Learning offers its tools (Blue, Red, Green tool in Runtime Workspace) for VisionPro 10.0 to
be directly used in its QuickBuild as one of a VisionPro tool (DLRuntimeTool). See Integration with Cognex
VisionPro and Known Issues on page 3.
l Cognex Designer Support Termination
o Since VisionPro Deep Learning runtime workspace can be directly imported into VisionPro 10.0 QuickBuild
the integration with Cognex Designer (the Plug-in for Cognex Designer) is no longer supported from
Resolved Issues
l Fixed an issue that a plug-in that was built in Examples (C:\ProgramData\Cognex\VisionPro Deep
Learning\2.0\Examples\c#\Example.Deep.Learning.Studio.GUI.Plugin) is not shown on Add Plugin window of Plugin
Manager.
25
Release History

l Red Analyze High Detail
o 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
l Loss Inspector
o 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).
l F1 Score of Confusion Matrix for Red Analyze tools
o The F1 score of the confusion matrix is introduced to show the result summary of the confusion matrix in Red
Analyze tools. See Red Analyze Focused Supervised - Confusion Matrix for an example of the F1 score of
confusion matrix.
o 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.
l Auto Threshold for Red Analyze High Detail and Red Analyze Focused Supervised
o 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. See Red Analyze Focused Supervised - Neural Network
Processing (Inferencing) for an example of the Auto Threshold.
l Region List for Red Analyze tools
o 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.
l Overlay Checkbox
o 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 Blue Locate - Review Markings for an example of the
overlay checkbox for Blue Locate tool.
l Restore Parameters
o For all tools, the restore parameter button is introduced for an easier "roll-back" of the set of parameter values
in the Tool Parameters pane toward those that were used in the last training session. See Blue Locate -
Neural Network Training for an example of the restore parameters button for Blue Locate tool.
l 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).
26
Release History
Resolved Issues
l Fixed an issue in Parameter Search that some texts in Database Overview are overlapped to each other and
displayed incorrectly.
l Multiple GPUs are now fully supported efficiently for High Detail modes. See Configure Multiple GPUs for the details.
l Fixed a bug that the multiple additions of the same plugin with Manage Plugins.
l 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.
l Fixed a bug in Blue Read that the overlay of marking and label is not correctly displayed.
l Fixed a bug in Red Analyze that the area under ROC curve is different from AUC value in Database Overview - ROC
tab.
l 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.
l Fixed some naming inconsistencies in C:\ProgramData\Cognex\VisionPro Deep Learning\1.1\Examples\c#
o The assembly name for Example.Training.HDRed.Console is modified to the right name.
o The project file name "Example.ViDi.Suite.GUI.Plugin" is changed to
"Example.Deep.Learning.Studio.GUI.Plugin"
o The assembly name with "space" delimiter is changed to the one with "period/full-stop" delimiter.

l Support for NVIDIA GeForce® RTX™ 30 series.
o VisionPro Deep Learning 1.0.1 supports RTX 3090, RTX 3080 and RTX 3070 for all tasks.
o 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 Configure Multiple GPUs for more details.
o Currently, there are some known issues relating to NVIDIA driver versions and Processing Time. See Known
Issues on page 3 for more details.
l Support for Optimized GPU Memory in Parameter Search
o If you use Focused Mode in Parameter Search, you can set and use Optimized GPU Memory.
o If you use High Detail Mode in Parameter Search, Optimized GPU Memory is automatically disabled for the
efficiency of High Detail Mode.
Resolved Issues
l Resolved an issue that the prediction result of a cloned Red Analyze Tool is incorrect when Low Precision is enabled.
l Resolved an issue that labeling is mistakenly applied on images, not views, after receiving views from an upstream
tool.
27
Release History
l Resolved an issue that the space between "Image Record" blocks you from using ImageRecord feature with
DeepLearningRuntimeBlock inside Designer.

l Two modes for the Green Classify Tool.
o High Detail mode (from SuaKIT), which provides higher classification accuracy on many challenging
applications.
o Focused mode (from ViDi) for simpler applications that need faster training or runtime execution.
o A Heat Map for High Detail mode, which visually indicates which portions of the image were most influential in
the classification decision.
o Validation sets for High Detail mode. These provide feedback on model performance and are used for training
optimization.
o You can enable/disable GPU memory optimization for use with High Detail mode and other tools.
l Simpler labeling controls, including
o Selecting multiple samples when label or accepting views, clearing markings and labels, processing, or
managing image sets.
o Automatically re-using the last label name by default, rather than requiring you to explicitly type each new
label.
Resolved Issues
l Resolved an issue with feature size indicator does not change after changing the value of the feature size in the
sampling tool parameter.
l Resolved an issue with "Save new version" in the workspace list does not act in the Blue Read Tool.
l Resolved an issue with name of Set are shown as original name in the image viewer even though users change the
name.
VisionPro ViDi 4.1.0

l Additional Blue Locate tool enhancements, including the following:
o New handles on the feature graphic labels to make resizing easier.
o Select either rectangular or circular feature graphic labels.
o Display or hide the name of feature graphic labels.
o Copy and paste feature label graphics, as well as cloning a feature graphic label by clicking on it.
o Auto-generate the creation of a layout model by selecting features and selecting Create Model.
l 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.
28
Release History
l 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
l 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.
l 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.

l Blue Locate tool
The Blue Locate tool has undergone significant overhaul, including the following:
o Create a Layout Model, which provides presence/absense verification of features in areas of an image.
o 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.
o Feature identifiers can now be up to 140 characters.
o Improved capabilities for oriented and scaled features.
o 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.
o Improved performance of node models, along with a new Model Editor GUI.
o The tool can now create a region of interest for downstream tools based on a feature, a match, or a node or
layout model.
l Training workflow
o 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.
l Image sample sets
o 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.
l Database Overview filtering
o While in Expert Mode, it is now possible to filter results and statistics in the Database Overview, using the
syntax of the Display filters.
l Blue Read tool
o Capability to identify and read rotated characters.
l Red Analyze tool
o An eraser tool has been added which allows you to remove areas of defect regions and masks.
l In addition to the above new features, there have been numerous performance and stability improvements.
29
Release History
Resolved Issues
l Resolved an issue where exporting or cloning a large workspace could cause the ViDi Suite GUI to become
unresponsive or unexpectedly quit.
l 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.
l Resolved an issue with the Blue Read tool's Label Views dialog not displaying the names of Regex models that had
been created.
l Resolved an issue with the Red Analyze tool's generated reports not displaying the markings in the report.

l General stability and performance improvements.
l Improved support for NVIDIA® Turing™ Tensor Cores for the Red Analyze and Green Classify tools with the Low
Precision parameter enabled.
l Two new model types have been added for the Blue Read tool.
o 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.
o A string model, which allows you to match linear strings based on the length of the text string.
l Improved performance for image displays that include a large number (500+) of feature graphics.
l Improved readability of the Green Classify tool's class labels when creating reports for the tool.
l Simplified controls for rotating a tool's Region of Interest (ROI).
l Support for Cognex VisionPro version 9.5 has been added.
Resolved Issues
Issue Description
AIC- Corrected an issue with NVIDIA drivers 411.31 and newer. For more information, please visit the Cognex Support
1861 Site VisionPro ViDi Software Driver Support page.
AIC- The Set Group option has been restored to the Actions for N Views menu in the View Browser, which allows you to
1801 create a group of images when using the Red Analyze tool in Unsupervised Mode.
AIC- Corrected an issue where a workspace would not function correctly if a stream contained a single quote in the
1201 name.
AIC- Corrected an issue where the ViDi Suite GUI would become unresponsive when applying a region for a Red
1223 Analyze tool with numerous views.
AIC- Corrected an issue where the ViDi Suite GUI would not terminate properly after attempting to connect to a runtime-
1387 only service.
30
Release History

l 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.
l This functionality supports up to 4 clients for deployment operation, and up to 8 clients for development.
l This release provides enhancements and speed improvements for the ViDi runtime:
o Support for NVIDIA® Tensor Cores has been added, and the GPU Optimization feature also now supports
Tensor Cores.
o The Low Precision parameter has been added to the Red Analyze and Green Classify tools, which optimizes
the tools for speed during runtime operation.
l 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.
l 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.
l Support for up to 32MP images (up to 16k in height or width) has been added.
Tool and GUI Changes

l The ViDi Blue Read Tool has been added for Optical Character Recognition (OCR) applications.
l Speed improvements and increased performance for the Blue Locate Tool when matching models.
l The Feature Size parameter is now interactive and can be graphically resized within the GUI.
l Labeling with the Blue tools has been improved to include multi-feature selection and interaction.
l The Blue tools have improved flexibility for model optimization.
l In the results, a Confusion Matrix has been added for the Blue Locate and Blue Read tools.
l A new feature/match and region-focus display mode has been added when using the Database Explorer.
l Checks have been added to provide feedback on wrongly labeled views.
l The examples have been updated, and there is a new example for the Blue Read tool.
l 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).
l The Red Analyze Tool – Statistics have been enhanced for better readability.
l There have been numerous issues resolved, and additional enhancements and improvements to the ViDi software.
31
Release History
Processing and Support Changes

l 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.
l 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.
l There is a new Cognex VisionPro wrapper assembly, which provides data conversion between VisionPro and ViDi
Suite, offering the following:
o Efficient image conversion
o Combined graphics
o Fixturing conversion
o Calibration conversion
l 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.
l Support for Cognex VisionPro 9.3 has been added.
l Support for Microsoft® Windows® 10 Professional (64-bit).
l Support for Microsoft® Windows® 7 Professional (64-bit).
l The performance when loading training Workspaces, utilizing on-demand loading of tools, has been improved.
l A new Workspace archive format, based on ZIP files, has been implemented.
32
API Changelog
API Changelog
This topic lists the API changes in previous releases of VisionPro Deep Learning.
GPU Clock Stabilizer

First, you can try adjusting the GPU clock settings to prevent lowering the GPU clock. Please change the 'Power
management mode' in Nvidia Control Panel > Manage 3D settings to 'Prefer maximum performance. However, if the GPU
clock still cannot be maintained, consider using together with the new feature called GPU Clock Stabilizer.
The feature to maintain a stable GPU clock in a deployment environment can be helpful in multi-GPU scenarios or in
situations where clock instability leads to reduced processing speed or spikes.
For more details, see the below example codes.
Note:
l This feature is initially disabled by default. You can control its activation or deactivation through the API.
l It should be employed after Compute Device initialization. Attempting to use it before initialization will result in
the C API returning a value other than VIDI_SUCCESS, while the .NET API will trigger an exception.
C++ Example of GPU Clock Stabilizer
1 ...
2 /** @brief disable stabilize feature */
3 #define VIDI_STABILIZE_OFF 0
4
5 /** @brief enable stabilize feature on GPUs */
6 #define VIDI_STABILIZE_GPU 1
7
8 /** @brief stabilize the compute devices
9 * @param mode mode of operation
10 * @return 0 if passed, otherwise an error_code that can be
used with vidi_get_error_message()
11 *
12 * @see VIDI_STABILIZE_GPU, VIDI_STABILIZE_OFF
13 * This method must be called after the compute devices have
been initialized.
14 */
15 VIDI_DLLEXPORT VIDI_UINT vidi_stabilize_compute_device(VIDI_
INT mode);
16
17 ///////////////////////////////////////////////////////
33
API Changelog
18 // example
19 ///////////////////////////////////////////////////////
20
21 // Turns on.
22 vidi_stabilize_compute_device(VIDI_STABILIZE_GPU);
23
24 // Turns off
25 vidi_stabilize_compute_device(VIDI_STABILIZE_OFF);
26 ...
C# Example of GPU Clock Stabilizer
1 ...
2 /// <summary>
3 /// Offers the possibility to stabilize the compute
devices when the control is created, without unloading the vidi dll
4 /// example: mode = "StabilizeMode.Off"
5 /// mode = "StabilizeMode.GPU"
6 /// This can only be called after compute devices
initialized.
7 /// </summary>
8 void StabilizeComputeDevices(Enum mode);
9
10
11 ///////////////////////////////////////////////////////
12 // example
13 ///////////////////////////////////////////////////////
14
15 // Turns on
16 control.StabilizeComputeDevices(StabilizeMode.GPU);
17
18 // Turns off
19 control.StabilizeComputeDevices(StabilizeMode.Off);
20 ...
vidi_initialize2() is Deprecated
vidi_initialize2() function has been deprecated.
Starting from version VisionPro Deep Learning 3.2.0, you can load and use both the training API and runtime API
simultaneously. Although the vidi_initialize2() API still exists, it does not perform the intended operation.
34
API Changelog
Deprecated vidi_initialize2()
1 VIDI_DLLEXPORT VIDI_UINT vidi_initialize2(VIDI_INT compute_mode, VIDI_STRING

compute_devices, VIDI_INT cuda_load_mode);

Among the new features introduced in VisionPro Deep Learning 3.0, some of them are only provided through API as they
enhance runtime functionality.
Classification Batch Processing for Runtime API

Processing multiple views at once through batch processing is now available for Green Classify High Detail. Using the
following Runtime APIs can speed up the Green Classify High Detail processing via API.
The steps to use classification batch processing API are summarized as follows.
1. Fix the batch size and prepare the images to process. The maximum batch size is dependent on the current available
GPU memory.
2. Create a sample.
3. Add the prepared images to the sample.
4. Run batch processing.
5. Check the processing results.
For more details of utilizing batch processing API, see the below example codes.
Note: If you select a batch size larger than the maximum batch size which is calculated based on the available memory
of the current GPU in Step 1, the batch processing will fail to start.
C++ Example of Batch Processing API for Runtime (Green High Detail)
1 ...
2 // The maximum batch size depends on the available GPU memory size.
3 // let's suppose the available maximum batch size is 12.
4 constexpr int batch_size = 12;
5
6 // you need to set "runtime_parameters/batch_size" as the batch size you
chose above.
7 // "workspace": the name of your workspace
8 // "default": the name of the stream in your workspace
9 // "Classify": the name of the Green High Detail tool in your stream.
35
API Changelog
10 string batch_size_str = std::to_string(batch_size);

11 vidi_runtime_tool_set_parameter("workspace", "default", "Classify", "runtime_
parameters/batch_size", batch_size_str.c_str());
12 ...
13 // you also need to prepare the same number of the images as the batch size
14 vector<VIDI_IMAGE> images(batch_size);
15 for (int i = 0; i < batch_size; i++)
16 {
17 ...
18 status = vidi_load_image(image_path.c_str(), &images[i]);
19 ...
20 }
21 ...
22 // create a sample for batch processing. Note that a sample is the basic unit
of processing task.
23 // "my_sample": the name of the sample you create.
24 status = vidi_runtime_create_batched_sample("workspace", "default", "my_
sample");
25 ...
26 // add the prepared images to the sample
27 for (int i = 0; i < batch_size; i++)
28 {
29 ...
30 status = vidi_runtime_batched_sample_add_image
("workspace", "default", "my_sample", &images[i]);
31 ...
32 }
33 ...
34 // execute batch processing
35 status = vidi_runtime_batched_sample_process
("workspace", "default", "Classify", "my_sample", "");
36 ...
37 // get result of batch processing
38 status = vidi_runtime_get_batched_sample("workspace", "default", "my_
sample", &result_buffer);
39 ...
C# Example of Batch Processing API for Runtime (Green High Detail)
...
// The maximum batch size depends on the available GPU memory size.
// let's suppose the available maximum batch size is 12.
int batch_size = 12;
...
36
API Changelog
// you also need to prepare the same number of the images as the batch size
// the paths list should contain the paths of images
List<IImage> imgs = new List<IImage>();
for (int i = 0; i < batch_size; ++i)
imgs.Add(new LibraryImage(paths[i]));
// create a sample for batch processing. Note that a sample is the basic unit of
processing task.
// "Classify": the name of the Green High Detail tool in your stream.
using (IBatchedSample sample = stream.CreateBatchedSample())
{
// add images to sample
sample.AddImage(imgs[i]);
ITool greenTool = stream.Tools["Classify"];
// set batch size parameter

var param = greenTool.ParametersBase as
ViDi2.Runtime.IGreenHighDetailParameters;
param.BatchSize = batch_size;
// run processing
sample.Process(greenTool);
// get processing result of batch

{
var tags = sample.Tags(greenTool, i);
foreach (var tag in tags)
Console.WriteLine($"tag: name={tag.Key}, score={tag.Value}");
}
}
Note: Batch processing should not be used in a tool chain. If you want to use batch processing in a chain of tools, create
a runtime workspace for each tool and chain them.
NVIDIA TensorRT Support for Runtime API

NVIDIA TensorRT is an NVIDIA SDK that boosts the inference speed of a deep learning application. It optimizes the
inference speed of a deep neural network by NVIDIA GPU model. From VisionPro Deep Learning 3.0, Runtime API supports
TensorRT to boost the processing speed at runtime. To exploit TensorRT for Runtime API, go through the following steps.
The TensorRT for Runtime API is supported only for Red Analyze High Detail and Green Classify High Detail runtime
environment.
37
API Changelog
1. In the training environment, create a workspace, stream, and a High Detail tool. Train the tool to make it ready to be
deployed in the runtime environment. The training environment can be either VisionPro Deep Learning GUI or API.
You can create a workspace, stream, and a High Detail tool and train the tool in GUI. For the more details of the
training and runtime environment, see Environments.
2. Export a runtime workspace which includes the trained High Detail tool. You also can create a runtime workspace in
GUI (Workspace > Export Runtime Workspace). For the more details of the runtime workspace, see Runtime
Workspace.
3. Deploy the runtime workspace and its High Detail tool on the device on your front line. For the more details of the
runtime deployment, see Runtime Workspace and Runtime Deployment.
38
API Changelog
4. Run TensorRT Optimization API to optimize the High Detail tools in your runtime workspace with TensorRT, and save
your runtime workspace. The optimization normally takes under 10 minutes to be completed. After the first run of
optimization, the tool's neural network is optimized to the specific NVIDIA GPU model you use, and from this point,
you can process as many images you want with faster speed.
C++ Example of TensorRT Optimization API for Runtime (Red High Detail and Green High Detail)
1 ...
2 // open the given workspace.
3 // "workspace": the name of your runtime workspace
4 //"..\\..\\resources\\runtime\\Green High-detail Tool.vrws": the path to the
runtime workspace file
5 status = vidi_runtime_open_workspace_from_file("workspace",
"..\\..\\resources\\runtime\\Green High-detail Tool.vrws");
6 ...
7
8 // optimize the High-detail tool
9 // "default": the name of the stream in your workspace
11 clog << "Start optimization. It will take a few minutes." << endl;
12 status = vidi_runtime_tool_convert_trt
("workspace", "default", "Classify", 0);
13 ...
14
15 // save runtime workspace with the optimized tool
16 string save_path = "..\\..\\resources\\runtime\\Green High-detail Tool
optimized.vrws";
17 status = vidi_runtime_save_workspace("workspace", save_path.c_str());
18 clog << "The workspace with the optimized tool is saved at " << save_path.c_
str() << endl;
19 ...
C# Example of TensorRT Optimization API for Runtime (Red High Detail and Green High Detail)
1 ...
2 // Open a runtime workspace from file
3 // the path to this file relative to the example root folder
4 // and assumes the resource archive was extracted there.
5 // "workspace": the name of your runtime workspace
6 ViDi2.Runtime.IWorkspace workspace = control.Workspaces.Add("workspace",
"..\\..\\..\\..\\resources\\runtime\\Green High-detail Tool.vrws");
7
39
API Changelog
8 // Store a reference to the stream 'default'

9 IStream stream = workspace.Streams["default"];
10
12 var greenTool = stream.Tools["Classify"] as ViDi2.Runtime.IGreenTool;
13
14 // Optimizes the High-Detail tool with TensorRT. It takes some time to
optimize the tool.
15 greenTool.OptimizeTensorRT(0);
16
17 // Save runtime workspace with the optimized tool.
18 // You can use this workspace to process with the optimized tool.
19 string savePath = "..\\..\\..\\..\\resource\\runtime\\Green High-detail Tool
Optimized.vrws";
20 workspace.Save(savePath);
21 Console.Write($"The workspace with the optimized tool is saved at
{savePath}.");
22 ...
If you want to use the Classification Batch Processing for Runtime API on page 35 together with TensorRT-optimized
processing, you need to set batch size before calling the optimization.
C++ Example of Setting Batch Size Before Calling TensorRT Optimization (Green High Detail)
1 // Optimize with batchSize=16

2 // The 5th(width) and 6th(height) parameter of vidi_runtime_tool_convert_trt()
are for target width and height for optimization
3 // If width=0 or height=0 is provided, the trained size is used. So we just
recommend use width=0, height=0 in normal case.
4 int gpu_index = 0;
5 int batch_size = 16;
6 vidi_runtime_tool_convert_trt("workspace", "default", "Analyze", gpu_
index, 0, 0, batch_size);
C# Example of Setting Batch Size Before Calling TensorRT Optimization (Green High Detail)
1 // Optimize with batchSize=16

2 // The second(width) and third(height) parameter of OptimizeTensorRT() are for
target width and height for optimization
40
API Changelog
3 // If width=0 or height=0 is provided, the trained size is used. So we just

recommend use width=0, height=0 in normal case.
4 int gpuIndex = 0;
5 int batchSize = 16;
6 greenTool.OptimizeTensorRT(gpuIndex, 0, 0, batchSize);
Directory to C++ Full Example Codes:

C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c++/Example.Runtime.OptimizeHDTool
C:\ProgramData\Cognex\VisionPro Deep Learning\3.0\Examples\c++\Example.Runtime.HDGreen.Batched
Directory to C# Full Example Codes:

C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c#/Example.Runtime.OptimizeHDTool.Console
C:\ProgramData\Cognex\VisionPro Deep Learning\3.0\Examples\c#\Example.Runtime.HDGreen.Batched.Console
41
API Changelog
5. Load your runtime workspace in your application and deploy it on the device to process images with the TensorRT-
optimized tool. To do this, you need to set a parameter before processing.
C++ Example of Setting Parameter to Use TensorRT Optimized-model for Processing (Red High Detail
and Green High Detail)
1 // If you want to process with optimized tool, you have to set ProcessTensorRT
to true.
2 // Setting this option to true will trigger TensorRT processing.
3 status = vidi_runtime_tool_set_parameter
("workspace", "default", "Analyze", "runtime_parameters/process_with_
trt", "true");
4 ...
5 status = vidi_runtime_sample_process("workspace", "default", "Analyze", "my_
sample", "");
C# Example of Setting Parameter to Use TensorRT Optimized-model for Processing (Red High Detail
and Green High Detail)
1 // If you want to process with optimized tool, you have to set

ProcessTensorRT to true.
2 // Setting this option to true will trigger TensorRT processing.
3 var hdParam = hdTool.ParametersBase as
ViDi2.Runtime.IToolParametersHighDetail;
4 hdParam.ProcessTensorRT = true;
5 ...
6 sample.Process(hdTool);
See the example codes below to process images with a runtime workspace.
Directory to C# Example Codes:

l C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c#/Example.Runtime.HDGreen.Console
l C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c#/Example.Runtime.HDRed.Console
Directory to C++ Example Codes:

l C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c++/Example.Runtime.HDGreen
l C:/ProgramData/Cognex/VisionPro Deep Learning/3.0/Examples/c++/Example.Runtime.HDRed
Note: TensorRT Optimization API (vidi_runtime_tool_convert_trt, .OptimizeTensorRT) should

be called again after your NVIDIA GPU model is changed because its optimization works by NVIDIA GPU
model. If you changed your NVIDIA GPU model in your device, repeat the Step 4 and 5.
42
API Changelog

In VisionPro Deep Learning 2.1.1, a new processing parameter max_scan_iterations is added only in API for Blue
Locate and Blue Read. This parameter limits the maximum number of iterations for image scanning during processing and
providing a fixed number for this parameter speeds up the processing.
Note: The features or the character features can be less detected than expected when the value of max_scan_
iterations is set too low. The appropriate value for this parameter is dependent on the images.
You can get the value of this parameter by:

l vidi_runtime_get_parameter
l vidi_runtime_tool_get_parameter
You can change the value of this parameter by:

l vidi_runtime_set_parameter
l vidi_runtime_tool_set_parameter
The parameter path for max_scan_iterations is:

l processing/blue/max_scan_iterations
Examples
This topic introduces the code examples of max_scan_iterations parameter for C and .NET API
C API
1 VIDI_UINT status = vidi_runtime_tool_set_parameter

("WORKSPACE", "STREAM", "TOOL", "processing/blue/max_scan_iterations", "40");
.NET API
1 libraryAccess.SetToolParameter("WORKSPACE", "STREAM", "TOOL",

"processing/blue/max_scan_iterations", "40"); // libraryAccess is
ILibraryAccess
43
API Changelog

As High Detail Quick modes are added in VisionPro Deep Learning 3.2.1, there are some changes to API as well.
Green Classify High Detail Quick Training - C API
Green High Detail Quick Training with C API
1 // Green high-detail-quick mode

2 ...
3 param_ss << "workspaces/" << workspace_name << "/streams/" << stream_name
<< "/tools/" << tool_name << "/tool_type";
4 status = vidi_training_set_parameter(param_ss.str().c_str(), "green_quick");
5 ...
6
To train Green Classify High Detail Quick, the parameter path should be set to "green_quick" for vidi_training_set_
parameter.
The example of parameter path:

l workspaces/WORKSPACE_NAME/streams/STREAM_NAME/tools/TOOL_NAME/tool_type
To configure the Tool Parameters for Green Classify High Detail Quick, the parameter path also should be set. For example,
if you want to change Epoch Count in Training Parameters, set the parameter path as:
l workspaces/WORKSPACE_NAME/streams/STREAM_NAME/tools/TOOL_NAME/training_parameters/count_
epochs
See C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c++\Example.Training.HDGreen\example_training_

hdgreen.cpp for the more detailed examples of training C API codes.
Green Classify High Detail Quick Training - .NET API
Green High Detail Quick Training with .NET API
1 ...
2 // Green high-detail-quick mode
3 hdGreenTool.Database.Tool.Type = ToolType.GreenQuickHighDetail;
4 ...
5
44
API Changelog
To train Green Classify High Detail Quick, the ViDi2.Training.ITool.Type should be set to
ToolType.GreenQuickHighDetail.
To configure the Tool Parameters for Green Classify High Detail Quick, the
ViDi2.Training.IGreenHighDetailParameters should be set. For example, you should set the value of
CountEpochs for IGreenHighDetailParameters.
See C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c#\Example.Training.HDGreen.Console\Program.cs
for the more detailed examples of training .NET API codes.
Green Classify High Detail Quick Processing - C API

The way to process Green Classify High Detail Quick with API is the same as those of Green Classify Focused and Green
Classify High Detail. See C:\ProgramData\Cognex\VisionPro Deep
Learning\2.1\Examples\c++\Example.Runtime.HDGreen\example_runtime_hdgreen.cpp for the details.
Green Classify High Detail Quick Processing - .NET API

The way to process Green Classify High Detail Quick with API is the same as those of Green Classify Focused and Green
Classify High Detail. See C:\ProgramData\Cognex\VisionPro Deep
Learning\2.1\Examples\c#\Example.Runtime.HDGreen.Console\Program.cs for the details.
Red Analyze High Detail Quick Training - C API
Red High Detail Quick Training with C API
1 ...
2 // Red high-detail quick mode
3 param_ss << "workspaces/" << workspace_name << "/streams/" << stream_name
<< "/tools/" << tool_name << "/tool_type";
4 status = vidi_training_set_parameter(param_ss.str().c_str(), "red_quick");
5 ...
6
To train Red Analyze High Detail Quick, the parameter path should be set to "red_quick" for vidi_training_set_
parameter.
The example of parameter path:

l workspaces/WORKSPACE_NAME/streams/STREAM_NAME/tools/TOOL_NAME/tool_type
45
API Changelog
To configure the Tool Parameters for Red Analyze High Detail Quick, the parameter path also should be set. For example, if
you want to change Epoch Count in Training Parameters, set the parameter path as:
l workspaces/WORKSPACE_NAME/streams/STREAM_NAME/tools/TOOL_NAME/training_parameters/count_
epochs
See C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c++\Example.Training.HDRed\example_training_

hdred.cpp for the more detailed examples of training C API codes.
Red Analyze High Detail Quick Training - .NET API
Red High Detail Quick Training with .NET API
1 ...
2 // Red high-detail quick mode
3 hdRedTool.Database.Tool.Type = ToolType.RedQuickHighDetail;
4 ...
To train Red Analyze High Detail Quick, the ViDi2.Training.ITool.Type should be set to
ToolType.RedQuickHighDetail.
To configure the Tool Parameters for Red Analyze High Detail Quick, the
ViDi2.Training.IRedHighDetailParameters should be set. For example, you should set the value of
CountEpochs for IRedHighDetailParameters.
See C:\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\c#\Example.Training.HDRed.Console\Program.cs for
the more detailed examples of training .NET API codes.
Red Analyze High Detail Quick Processing - C API

The way to process Red Analyze High Detail Quick with API is the same as those of Red Analyze Focused Supervised and
Red Analyze High Detail. See C:\ProgramData\Cognex\VisionPro Deep
Learning\2.1\Examples\c++\Example.Runtime.Red\example_runtime_red.cpp for the details.
Red Analyze High Detail Quick Processing - .NET API

The way to process Red Analyze High Detail Quick with API is the same as those of Red Analyze Focused Supervised and
Red Analyze High Detail. See C:\ProgramData\Cognex\VisionPro Deep
Learning\2.1\Examples\c#\Example.Runtime.HDRed.Console\Program.cs for the details.
46
API Known Issues
API Known Issues

l For the adaptation task, it is unable to build its API example codes in Visual Studio 2015. Use Visual Studio 2017 or
2019 instead to build them and run the examples. See Adaptation Mode: Adapting to Line Variation for more details of
the adaptation.
l With API, heatmap is not supported in Green Classify High Detail. For more details, please refer to the API
documentation.
l There is an issue where an out of memory error may occur when using a Runtime Workspace from versions prior to
VisionPro Deep Learning 3.2 after TensorRT optimization in VisionPro Deep Learning 3.2 during API processing,
even when the image sizes are the same.
47
API Known Issues
48
API Deprecated Features

The following list of C and .NET functions are being deprecated, and will either be renamed or removed in the next major
release.
Library Obsolete Name New Name

.NET ToolDatabase.GetOverlayImage()
.NET ViewKey.Filename FrameKey.SampleName
.NET LibraryAccess.SampleGetOverlayImage()
.NET IManualRegionOfInterest.External IManualRegionOfInterest.Parameters.Mode
.NET IManualRegionOfInterest.Mode IManualRegionOfInterest.Parameters.Mode
.NET IManualRegionOfInterest.Units IManualRegionOfInterest.Parameters.Units
.NET IManualRegionOfInterest.Offset IManualRegionOfInterest.Parameters.Offset
.NET IManualRegionOfInterest.Size IManualRegionOfInterest.Parameters.Size
.NET IManualRegionOfInterest.Angle IManualRegionOfInterest.Parameters.Angle
.NET IManualRegionOfInterest.SplittingGrid IManualRegionOfInterest.Parameters.SplittingGrid
.NET IBlueRegionOfInterest.Model IBlueRegionOfInterest.Parameters.MatchFilter
.NET IBlueRegionOfInterest.Models
.NET IBlueRegionOfInterest.TargetMode IBlueRegionOfInterest.Parameters.TargetMode
.NET IBlueRegionOfInterest.FeatureFilter IBlueRegionOfInterest.Parameters.FeatureFilter
.NET IBlueRegionOfInterest.MatchFilter IBlueRegionOfInterest.Parameters.MatchFilter
.NET IBlueRegionOfInterest.MatchItems IBlueRegionOfInterest.Parameters.MatchItems
.NET IBlueRegionOfInterest.MatchItemsMode IBlueRegionOfInterest.Parameters.MatchItemsMode
.NET IBlueRegionOfInterest.Centered IBlueRegionOfInterest.Parameters.Centered
.NET IBlueRegionOfInterest.Units IBlueRegionOfInterest.Parameters.Units
.NET IBlueRegionOfInterest.Offset IBlueRegionOfInterest.Parameters.Offset
.NET IBlueRegionOfInterest.Size IBlueRegionOfInterest.Parameters.Size
.NET IBlueRegionOfInterest.Angle IBlueRegionOfInterest.Parameters.Angle
.NET IBlueRegionOfInterest.Scaled IBlueRegionOfInterest.Parameters.Scaled
.NET IBlueRegionOfInterest.Scale IBlueRegionOfInterest.Parameters.Scale
.NET IRedRegionOfInterest.ExtractRegions IRedRegionOfInterest.Parameters.ExtractRegions
.NET IRedRegionOfInterest.RegionFilter IRedRegionOfInterest.Parameters.RegionFilter
.NET IRedRegionOfInterest.Units IRedRegionOfInterest.Parameters.Units
.NET IRedRegionOfInterest.Border IRedRegionOfInterest.Parameters.Border
.NET IRedRegionOfInterest.MaskRegions IRedRegionOfInterest.Parameters.MaskRegions
.NET IRedRegionOfInterest.FailOnFilter IRedRegionOfInterest.Parameters.FailOnFilter
.NET IRedRegionOfInterest.RecycleMask IRedRegionOfInterest.Parameters.RecycleMask
.NET IRedRegionOfInterest.InvertMask IRedRegionOfInterest.Parameters.InvertMask
49

.NET IRedRegionOfInterest.GrowShrinkMask IRedRegionOfInterest.Parameters.GrowShrinkMask
.NET IGreenRegionOfInterest.Filter IGreenRegionOfInterest.Parameters.Filter
.NET IMatch.FeatureString IReadModelMatch.FeatureString
.NET IMatch.PrettifiedString IReadModelMatch.PrettifiedString
.NET IMatch.Points INodeModelMatch.NodePoints
.NET IMatch.Pose INodeModelMatch.Pose
.NET IMatch.Position IReadModelMatch.Position
.NET IMatch.Scale INodeModelMatch.Scale
.NET IMatch.AspectRatio INodeModelMatch.AspectRatio
.NET IMatch.Shear INodeModelMatch.Shear
.NET IModel.Occurences IModel.Occurrences
.NET INode.Names INode.Fielding
.NET ITool.Parameters ITool.ParameterBase
.NET IFeature.Id IFeature.Name
.NET IRedView.Deviation IRedView.Score
.NET ITag.Key ITag.Name
.NET ITag.Value ITag.Score
.NET IWorkspace.Name IWorkspace.UniqueName
.NET IWorkspace.LastModification
.NET IToolDatabase.SetTrainFlag(string filter, string group) IToolDatabase.SetTrainFlag(string filter, bool flag)
.NET IToolDatabase.GetOverlayImage(ViewKey, string, uint)
.NET ILibraryAccess.ToolGetOverlay(WorkspaceName,
StreamName, ToolName, string, uint, string,
ILibraryAccessContext)
.NET IStream.AddImage(Image, string)
IStream.AddImage(System.IO.Stream, string)
GetImage(string, uint)
.NET ITool.Mask ITool.RegionOfInterest.Mask
.NET ITool.Parameters ITool.ParametersBase
.NET IToolParameters.TrainSelection
.NET IToolParameters.KeepTrainSelection
.NET BorderType.White
BorderType.Reflect
BorderType.Wrap
.NET Node.Names Node.Fielding
Node.IDs
.NET ProductCode.OptimizedMemory ProductCode.Standard
ProductCode.Advanced
50

.NET CalibratedRedView.Deviation
.NET TrainingFlagMode.Auto
.NET RedTool.IRedTool.Parameters RedTool.IRedTool.BaseParameters
C vidi_runtime_init_sample() vidi_runtime_create_sample()
C vidi_runtime_process() vidi_runtime_sample_process()
C vidi_runtime_process_sample() vidi_runtime_sample_process()
C vidi_runtime_get_image() vidi_runtime_sample_get_image()
C vidi_runtime_sample_get_overlay() Use get_overlay for the specific tool.
C vidi_training_tool_get_status() vidi_training_stream_list_tools()
C vidi_training_stream_process_sample() vidi_training_sample_process()
C vidi_training_blue_add_feature() vidi_training_set_feature()
C vidi_training_blue_add_model() vidi_training_blue_create_model()
C vidi_training_tool_set_mask() vidi_training_roi_set_mask()
C vidi_training_tool_get_mask() vidi_training_roi_get_mask()
C vidi_training_init_sample() vidi_training_add_sample()
C vidi_training_process_sample() vidi_training_sample_process()
C vidi_training_sample_get_overlay()
C vidi_training_tool_get_overlay()
C occurences occurrences
C names fielding
C red/mode training/network_model
C vidi_training_tool_train() vidi_training_tool_train2()
C vidi_initialize2()
51
Documentation Updates
l Added "{}" to the example JSON file in the Cognex Label Import/Export Plug-in page.

l The document structure and contents organization is reformed for the clear understanding, quick and easy access to
the necessary contents, and future enhancement and rearrangement of the contents.
l Fixed a typo "8192x819" to "8192x8192" in Known Issues on page 3 - Tools - General on page 4 section
l Fixed contents into that the Expert Mode does not support Edit Mask and Edit Region in Red Analyze . For example,
see Edit Mask.
l Updates about Adaptation Modes
o Adaptation Mode: Adapting to Line Variation
o General Workflow
o Green Classify High Detail Mode Parameter
o Red Analyze High Detail Mode Parameter
l Updates About Label Check
o Optimize Labeling with Label Check
l Updates on Parameter Search and its Auto mode.
o Parameter Search
o Manual Mode
o Auto Mode
l Updates on VisionPro Deep Learning GUI refinement
o GUI and API
o Interpret Results (all tools)
l Updates on Red Analyze High Detail Validation Loss Formula
o Validation Set and Validation Loss
l Updates about Indicators (Training Set Indicator, Validation Set Indicator) on Image Display Area, Updates about
Filters for Validation Set
o GUI Overview
o Prepare Validation Set (Red Analyze High Detail)
o Prepare Validation Set (Green Classify High Detail)
o Default Display Filters
o Tool-Specific View-Level Variables
52
l Updates about View Browser List Mode for all tools

o Define Region of Interest (ROI) and Mask to Construct View (all tools)
o Collect Images (all tools)
o Create an Image Sample Set (all tools)
o Modify Training Set (all tools)
o Review Markings (Red Analyze, Green Classify)
o Create Label (Labeling) (Red Analyze, Green Classify)
o View Browser - Grid Mode
o View Browser - List Mode
l Updates about TensorRT for Runtime API and Batch Processing for Runtime Green Classify
o API Changelog on page 33
l Updates about Cognex Label Import/Export Plugin
o Cognex Label Import/Export Plug-in
l Updates about Image Processing Logic Change for Red Analyze High Detail
o Configure Processing Parameters
l Updates about Minimum Image Size for Green Classify High Detail and Red Analyze High Detail.
l Collect Images (Green Classify High Detail)
l Collect Images (Red Analyze High Detail)
l Updates about Validation Loss Scale Change
o Validation Set and Validation Loss (Red Analyze High Detail)
o Validation Set and Validation Loss (Green Classify High Detail)
l Updates about Image Resize Improvement for Green Classify High Detail
o Resize
l Removing Red Analyze High Detail Quick Contents
o Removal of Red Analyze High Detail Quick on page 15
l Removing Stride parameter from Red Analyze High Detail Training parameters
o Training Parameters
l Updates about Migrating SuaKIT Segmentation and Classification Projects into VisionPro Deep Learning
o Import a SuaKIT Segmentation Project
o Import a SuaKIT Classification Project
l Updates on FAQ
l Updates on VisionPro Deep Learning 3.0 Known Issues
o Tools - Red Analyze
o Tools - Green Classify
o Parameter Search
o GPU
53
l Updates on VisionPro Deep Learning 2.1 Known Issues

o Parameter Search
l Updates on High Detail training logic regarding Patience Epochs and Minimum Epochs in Training parameters.
o Training Parameter Details: Patience Epochs & Minimum Epochs
l Updates on Best Epoch update logic in Loss Inspector
o Status Variables
l Updates on the compatible VisionPro version
o Integration with Cognex VisionPro
l Updates on the GPU driver versions for NVIDIA RTX / Quadro® / Data Center driver
o Pre-requisites: NVIDIA® GPU Driver
l Updates on Designer Plug-in support
o Using Runtime Workspace with Cognex Designer

l Updates on Red Analyze Tool and Green Classify Tool for newly added High Detail Quick modes
l Updates on Edit Deployment Target page that introduces the CIE1 deployment is no longer available
l Updates on Parameter Search and Use the Parameter Search Utility
l Updates on Image File and Format Requirements.
l Updates on the contents explaining statistical results
o Statistics
o False Positive and False Negative Results
o Blue Locate Tool – Statistics
o Blue Read Tool – Statistics
o Green Classify Tool – Confusion Matrix
o Green Classify Tool – F Score
o Red Analyze Tool – Statistics – Confusion Matrix
o Red Analyze Tool – Statistics – Region Area Metrics
l Updates on the model scores for Blue Locate/Read
o Blue Locate Tool – Node Model
o Blue Locate Tool – Layout Model
o Blue Read Tool – Node Model
o Blue Read Tool – Regex Model
o Blue Read Tool – String Model
54
l Updates on C#, C++ example codes (...\ProgramData\Cognex\VisionPro Deep Learning\2.1\Examples\). There were
many improvements in example codes and some of the notable changes are:
o The unit of angle for input is changed from "degree" to "radian" (example: Perturbation of Red Analyze)
o For Green Classify, the codes that load label information per image is added.
o The parameter path used for vidi_runtime_get_parameter() and vidi_runtime_set_parameter() is changed to
be a "full path" which should include "tool/" within it.
Example: "workspaces/(workspace_name)/streams/(stream_name)/tools/(tool_
name)/tool/processing/sampling_density"
Note: When either "roi/", "runtime_parameters/", or "post_process_parameters/" are used in the

parameter path, this path should not include "tool/".
Example 1: "workspaces/(workspace_name)/streams/(stream_name)/tools/(tool_name)/roi/..."
Example 2: "workspaces/(workspace_name)/streams/(stream_name)/tools/(tool_name)/runtime_
parameters/..."
l Updates on page contents.

l Updates for VisionPro Deep Learning 2.0 release
o The documentation for the integration with Cognex Designer is no longer provided

l Updates for VisionPro Deep Learning 1.1 release
l Designer Plugin script examples for each tool have been updated for VisionPro Deep Learning 1.1.
l Updates of the comments in C#, C++ example codes (...\ProgramData\Cognex\VisionPro Deep
Learning\1.1\Examples\)
l Updates about how to use multiple GPUs for VisionPro Deep Learning in Multiple GPUs Utilization.
l Updates about the introduction to the various training parameters for VisionPro Deep Learning in Training
Parameters
l Updates about how to import SuaKIT projects in Import SuaKIT Project.

l Updates for VisionPro Deep Learning 1.0.1 release
l Updates for Migrating To This Release
The migration records from ViDi Suite 3.4 to ViDi Suite 4.1.0 have been reorganized for the correct migration history.
Note: You can access the latest VisionPro Deep Learning Help on the Cognex Support website:
https://support.cognex.com/en/documentation/deep-learning
55
Preparation
Preparation
This topic covers the computer operating systems, hardware requirements, dongle configuration, licenses, and legal notices
you need to be prepared with before using VisionPro Deep Learning for your application.
Pre-requisites: Operating System

Microsoft® Windows® Operating System English Chinese Japanese Korean French German Spanish
Windows 10 Professional (64-bit) ✓ ✓ ✓ ✓ ✓ ✓ ✓
Windows Server 2019 ✓ ✓ ✓ ✓ ✓ ✓ ✓
Note: The use of the VisionPro Deep Learning software or the Deep Learning Service on a virtual machine (VM) is not
supported.
Note: Due to the issues from the NVIDIA driver, Windows Server 2016 is not supported for VisionPro Deep
Learning3.2.1 Server/Client mode. See Known Issues on page 3 for more details and the workaround.
Pre-requisites: Training PC Requirements

The following requirements describe the suggested system for the VisionPro Deep Learning application training and
development PC.
l CPU
We recommend the higher specification than Intel Core i7. When selecting a CPU, a higher CPU clock speed rate
and multiple core processors will result in faster runtime tool execution. If your application will be relying on the Blue
Locate tool, it is more sensitive to clock speed rates, particularly with complex model matching applications.
l GPU
When selecting a GPU, Cognex only supports NVIDIA GPUs.
For training, we recommend GPUs that have 10GB or larger GPU Memory (1080 Ti, 2080 Ti, 3080). For the
processing of High Detail modes and High Detail Quick modes, GPUs that have 8GB or larger GPU Memory are
recommended.
l System Memory (RAM)
First, 32GB or more than 32GB is recommended.
When selecting the system RAM, specify the greater of:
o The sum of all GPU memory. For example, if you have four NVIDIA GeForce® RTX™ 3080 Ti GPUs, which
each have 10 GB of memory, the PC should have 40 GB of RAM.
o The RAM should be one and a half times the typical workspace size. For example, if your typical workspace is
20 GB, the PC should have a minimum of 30 GB of RAM.
Note: You should have more than 50GB to train the tool with 15,000 images of 1K X 1K.
56
Preparation
l System Storage
Cognex recommends the use of a solid-state drive (SSD), we recommend 100GB + as a free space.
l Power Supply
When selecting your power supply, include a 25% margin above the requirement to accommodate the system and
GPU power requirements, in other words, select a power supply that is 1.25 times the system and GPU power
requirements.
l USB 2.0
USB port for a permanent Cognex Security Dongle (containing the training license) connection via USB 2.0.
Note:
The VisionPro, Designer and VisionPro Deep Learning software require that a valid Cognex Security Dongle be
Note: When VisionPro Deep Learning is configured for the Client/Server functionality, and a computer has been
Pre-requisites: Deployment PC Requirements

The following requirements describe the suggested system for the deployment of your VisionPro Deep Learning application
on a runtime PC.
l CPU
We recommend the higher specification than Intel Core i7. When selecting a CPU, a higher CPU clock speed rate
and multiple core processors will result in faster runtime tool execution. If your application will be relying on the Blue
Locate tool, it is more sensitive to clock speed rates, particularly with complex model matching applications.
For High Detail modes and High Detail Quick, if you have only CPU; not GPU, you cannot process the tool.
If a single core performs multi-tasks, processing may be jumped up. Therefore, it is necessary to use a sufficient
amount of resources or allocate the single task to a single core.
57
Preparation
l GPU
When selecting a GPU for processing, VisionPro Deep Learning3.2.1 only supports NVIDIA GPUs. Consider the
following when choosing an NVIDIA GPU:
At a minimum, an NVIDIA® CUDA® enabled GPU is required.
When selecting an NVIDIA GPU, a unit with a higher core clock frequency, CUDA cores and Tensor cores will result
in faster computational speeds. Due to this, Cognex strongly recommends the use of NVIDIA RTX / Quadro® and
Tesla GPUs for the following reasons:
o These GPUs support the compute-optimized Tesla Compute Cluster (TCC) mode.
o These GPUs are designed and tested for continuous duty-cycle operation.
o These GPUs undergo rigorous testing and qualification by NVIDIA.
In addition, these GPUs offer longer-term availability and driver stability.

8 GB of GPU memory is recommended for runtime. For the processing of High Detail modes and High Detail Quick
modes, GPUs that have 8GB or larger GPU Memory are recommended.
l System Memory (RAM)
We recommend 32GB or higher.
l PCIe Lanes
Cognex recommends a minimum of x8 PCIe lanes. However, a PCIe x16 has the potential to reduce cycle time by
approximately 10 ms, relative to a PCIe x8 (based on a 5 MP image).
l Power Supply
When selecting your power supply, include a 25% margin above the requirement to accommodate the system and
GPU power requirements, in other words, select a power supply that is 1.25 times the system and GPU power
requirements.
l USB 2.0
USB port for a permanent Cognex Security Dongle (containing the runtime license) connection via USB 2.0.
Note:
The VisionPro, Designer and VisionPro Deep Learning software require that a valid Cognex Security Dongle be
Pre-requisites: NVIDIA® GPU Device Requirements

The following information covers the requirements when utilizing an NVIDIA GPU with your VisionPro Deep Learning
application. The GPU is utilized by Deep Learning during the development of your application, typically with the training of
tools. In addition, a GPU can also be used during runtime deployment, where it increases performance of runtime
workspaces.
58
Preparation
VisionPro Deep Learning3.2.1 aims to support the most common GPU families, but it’s not practical to test every model or
driver. The table below indicates those models we have explicitly tested and confirmed work, as well as those few models
and families that do not work. Most other common NVIDIA GPUs other than very new models have been used successfully
by customers, even if they haven’t been explicitly tested. In addition, we share specifications for minimum GPU capability, as
well as recommendations high performance GPUs for demanding applications, such as those using Green Classify High
Detail or Red Analyze High Detail.
l Minimum recommended performance: GTX 1060 6GB

l Recommended for high performance: GTX 1080 Ti / RTX 2080 Ti / RTX 3080
Recommended for High

GPU Model Tested by Cognex
Performance
GTX 1060 6GB Tested
GTX 1070
GTX 1080 Tested
GTX 1080 Ti Tested Recommended
RTX™ 2070
RTX 2080
RTX 2080 Ti Tested Recommended
NVIDIA GeForce®
RTX 3060
RTX 3060 Ti
RTX 3070
RTX 3070 Ti
RTX 3080 Tested Recommended
RTX 3080 Ti Tested
RTX 3090 Tested Recommended
P2000
P4000 Tested
P5000 Recommended
GV100* Recommended
RTX 4000
NVIDIA RTX / Quadro®
RTX 5000 Recommended
Titan V* Recommended
Titan RTX Recommended
NVIDIA Tesla® V100*
59
Preparation
Recommended for High

GPU Model Tested by Cognex
Performance
Models known not to work:
l GTX 1660 and GTX 1650: Please note that low performance and slow training time with unknown reason.
Note:
l When selecting a GPU, Cognex only supports NVIDIA GPUs.
l VisionPro Deep Learning3.2.1 supports the use of any NVIDIA CUDA® enabled GPU that provides a compute
capability of 3.0 and higher, and requires a driver version greater than or equal to 528.02.
l CUDA compute capability 5.0 or higher is required to use the core VisionPro Deep Learning functionality. The
Low Precision parameter requires CUDA compute capability 6.1 or higher (for more information, visit the
NVIDIA website).
l When training and processing High Detail mode and High Detail Quick mode, a requirement of 8 GB of GPU
memory is for runtime operation, and 10 GB of GPU memory is recommended for training.
l The NVIDIA GeForce Titan V, NVIDIA RTX / Quadro® GV100 and Tesla V100 include NVIDIA Tensor Cores.
VisionPro Deep Learning3.2.1 will automatically utilize Tensor Cores if they are present, provided you have a
Standard or Advanced security license. Please contact your Cognex Sales Representative for information
about enabling the functionality on your Cognex Security Dongle.
Pre-requisites: NVIDIA® GPU Driver

l GeForce® RTX driver greater than or equal to 528.02
l NVIDIA RTX / Quadro® / Data Center driver greater than or equal to 528.02 (Optimal Driver for Enterprise).
o It is highly recommended to visit https://www.nvidia.com/Download/Find.aspx to find the right versions for
your GPU as your GPU may require higher version than 528.02.
Pre-requisites: API Development Requirements

The following software and components are necessary for developing VisionPro Deep Learning custom applications through
the VisionPro Deep Learning API:
l Microsoft® Visual Studio® 2015, 2017 or 2019
l Microsoft .NET Framework 4.7.2
Note: For more information about the VisionPro Deep Learning API, please refer to the Cognex Deep Learning .NET
Documentation and the Cognex Deep Learning C Documentation help files.
Pre-requisites: Image File and Format Requirements

The supported input image channels, bit depths, and image file formats in VisionPro Deep Learning and the image channels,
bit depths, and image file formats converted after importing to VisionPro Deep Learning are described in the table below. The
image channels, bit depths, and image file formats not described in this table are not supported in VisionPro Deep Learning.
60
Preparation
Before Importing into VisionPro Deep Learning After Importing into VisionPro Deep Learning
Input Input Input Imported Imported Imported

Image Channel Image Bit Depth Image File Format Image Channel Image Bit Depth Image File Format
1 ch 8-bit bmp 1 ch 8-bit png

1 ch 8-bit jpg 1 ch 8-bit png
1 ch 8-bit png 1 ch 8-bit png
1 ch 8-bit tiff 1 ch 8-bit png
3 ch 8-bit jpg 3 ch 8-bit png
Maximum Image Size and Dimensions

The following guidelines will help you determine how best to configure your Deep Learning application, within these
operational parameters. The figures in the table below (for example the maximum number of images per stream) are
experimentally tested and the values vary more or less on which hardware settings you rely. Generally, the larger your
memory (not GPU RAM) and hard disk capacity you have, the more the number and the size of the image you can handle.
l Maximum image size (Channel parameter set to 1 to 4)

o Images up to 32MP (up to 16k in height or width) are supported
o Images up to 32k pixels wide (in one dimension, dependent upon GPU resources)
o Images over 32k are not supported
l Maximum number of images per stream for Red Analyze
o 1000
l Maximum number of defects per image for Red Analyze
o 1000
Tip: As the number and the size of images are getting larger, the training and processing time for all images increases
as well.
61
Preparation
Install VisionPro Deep Learning

To successfully install VisionPro Deep Learning, perform the following:
1. Attach the Cognex Security Dongle to a USB port on the computer that will be used to develop the vision application.
2. Download the CognexVisionPro Deep Learning installer from the Cognex support page.
3. Run the VisionPro Deep Learning installer and follow the prompts.
Note:
Choosing the Custom option will allow you to install selected features:
l The Wibu Runtime Server, which is needed to connect to the USB Cognex Security Dongle.
l The main VisionPro Deep Learning application (this is required).
l The VisionPro Deep Learning Developer API.
l The VisionPro, Designer and VisionPro Deep Learning software require that a valid Cognex Security Dongle be
Launch VisionPro Deep Learning

When launching the Deep Learning GUI, there are several options that you can use to control the GPU Mode, which GPU
device to use and the allocation of GPU memory, in addition to project settings.
At the Starting Cognex Deep Learning screen, press the Options button.
62
Preparation
Command Settings
GPU Mode Specifies the GPU mode to be used by the application.
SingleDevicePerTool A single GPU is used for training and processing an image.
MultipleDevicesPerTool MultipleDevicesPerTool that aimed to provide processing an image on multiple
devices has no longer been supported since VisionPro Deep Learning 1.0.0.
NoSupport Specifies that a GPU will not be used.
Note: This option conflicts with --gpu-devices and/or --optimized-gpu-

memory.
Optimized Specifies the size of the pre-allocated optimized memory buffer. This setting is activated by default, with the
GPU default size of 2GB.
Memory
Tip: Compute Devices
You can set Optimized GPU Memory in the Help > Compute Devices option.
Compute Devices window contains the information about computing resources.
1. Available Devices
This shows the GPUs that are used for VisionPro Deep Learning. You can check list of multiple
devices and GRAM information here.
2. Optimized GPU Memory
You can enable/disable GPU memory optimization for the use of Focused mode tools. See
NVIDIA GPU Selection and Configuration for the details.
Auto Save Specifies how often a workspace will be auto-saved. The default is every 5 minutes.
Interval
Training Specifies a cache location for training images. This is useful if the images in a stream are in non-raster format
Image (for example JPG, PNG, or GIF), since they must be converted to a raster (BMP) format for training. By
Cache Path default, this conversion happens multiple times for each image. By enabling the Training Image Cache Path,
the conversion is only performed once and the converted images are stored in a local cache directory. The
cache directory should be local, and preferably on a Solid-State Drive (SSD).
This option can also speed up training in cases where the workspace is stored on a slower drive or a remote
storage device.
Locale Specifies the language to use throughout the VisionPro Deep Learning GUI.
Debug Specifies whether or not debug logs should be activated for the project.
Logging
CognexVisionPro Support
VisionPro Deep Learning is qualified to support the following CognexVisionPro release:
l VisionPro9.20
l VisionPro 9.8
For more information about the integration with CognexVisionPro, see Integration with Cognex VisionPro.
63
Legal Notices
Legal Notices
The software described in this document is furnished under license, and may be used or copied only in accordance with the
terms of such license and with the inclusion of the copyright notice shown on this page. Neither the software, this document,
nor any copies thereof may be provided to, or otherwise made available to, anyone other than the licensee. Title to, and
ownership of, this software remains with Cognex Corporation or its licensor. Cognex Corporation assumes no responsibility
for the use or reliability of its software on equipment that is not supplied by Cognex Corporation. Cognex Corporation makes
no warranties, either express or implied, regarding the described software, its merchantability, non-infringement or its fitness
for any particular purpose.
The information in this document is subject to change without notice and should not be construed as a commitment by
Cognex Corporation. Cognex Corporation is not responsible for any errors that may be present in either this document or the
associated software.
Companies, names, and data used in examples herein are fictitious unless otherwise noted. No part of this document may be
reproduced or transmitted in any form or by any means, electronic or mechanical, for any purpose, nor transferred to any
other media or language without the written permission of Cognex Corporation.
Copyright © 2024. Cognex Corporation. All Rights Reserved.
Portions of the hardware and software provided by Cognex may be covered by one or more U.S. and foreign patents, as well
as pending U.S. and foreign patents listed on the Cognex web site at: cognex.com/patents.
The following are registered trademarks of Cognex Corporation:

Cognex, 2DMAX, Advantage, AlignPlus, Assemblyplus, Check it with Checker, Checker, Cognex Vision for Industry,
Cognex VSOC, CVL, DataMan, DisplayInspect, DVT, EasyBuilder, Hotbars, IDMax, In-Sight, Laser Killer, MVS-8000,
OmniView, PatFind, PatFlex, PatInspect, PatMax, PatQuick, SensorView, SmartView, SmartAdvisor, SmartLearn,
UltraLight, Vision Solutions, VisionPro, VisionView
The following are trademarks of Cognex Corporation:
The Cognex logo, 1DMax, 3D-Locate, 3DMax, BGAII, CheckPoint, Cognex VSoC, CVC-1000, FFD, iLearn, In-Sight (design
insignia with cross-hairs), In-Sight 2000, InspectEdge, Inspection Designer, MVS, NotchMax, OCRMax, PatMax RedLine,
ProofRead, SmartSync, ProfilePlus, SmartDisplay, SmartSystem, SMD4, VisiFlex, Xpand
Portions copyright © Microsoft Corporation. All rights reserved.
Portions copyright © MadCap Software, Inc. All rights reserved.
Other product and company trademarks identified herein are the trademarks of their respective owners.
64

Cognex Deep Learning Release Notes

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Cognex Deep Learning Release Notes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cognex Deep Learning Release Notes

Uploaded by

Copyright:

Available Formats

Release Notes

VisionPro Deep Learning 3.2.1

VisionPro Deep Learning 3.2

Cognex Deep Learning Studio (GUI)

Tools - Red Analyze

Tools - Green Classify

Tools - Blue Read

VisionPro Deep Learning Service

VisionPro Deep Learning 2.1 (2.0, 2.1.1)

Operating System (OS)

Cognex Deep Learning Studio (GUI)

Tools - Red Analyze

Tools - Green Classify

Tools - Blue Read

VisionPro Deep Learning Service

Migration from SuaKIT or The Past VisionPro Deep Learning Versions

Integration with VisionPro

VisionPro Deep Learning 1.1 (1.1.1)

Operating System (OS)

Cognex Deep Learning Studio (GUI)

l Especially with bigger than 10GB workspaces,

Tools - Red Analyze

Tools - Green Classify

Tools - Blue Read

VisionPro Deep Learning Service

Migration from SuaKIT or The Past VisionPro Deep Learning Versions

Integration with VisionPro

VisionPro ViDi 3.1.0 ~ VisionPro Deep Learning 1.0.1

Operating System (OS)

Cognex Deep Learning Studio (GUI)

Tools - Red Analyze

Tools - Green Classify

Tools - Blue Read

VisionPro Deep Learning Service

Migration from SuaKIT or The Past VisionPro Deep Learning Versions

Integration with VisionPro

Migrating to This Release

Re-optimization Required When Using TensorRT

2.1 (2.1.1) → 3.0

Removal of "Extra Large" from Green Classify High Detail Network

Removal of Red Analyze High Detail Quick

High Detail Backwards Compatibility

l The trained network of this tool will be lost.

Runtime Workspace Backwards Compatibility

SuaKIT Segmentation and Classification Projects Compatibility

2.0 → 2.1 (2.1.1)

Termination of CIE1 Deployment Support

Termination of Designer Support

Change in Validation Loss Calculation

Loss Tracking in Log Messages

Region Filter for Confusion Matrix

VisionPro ViDi 4.1 → VisionPro Deep Learning 1.0

Architecture of Existing Green Classify Tool

Keeping Existing names of APIs

Internal version of APIs

No longer supported or Deprecated APIs

Changes to Designer functionality

Changes to the Feature Filter Syntax