Release Information

The VisionPro Deep Learning 3.3 introduces the following features and enhancements:

Features

• Improved Red Analyze and Green Classify High Detail tools

  Compared to the previous versions, the improved Red Analyze and Green Classify High Detail tools can be trained to the same level of performance faster. This feature affects the following tool parameters of the Red Analyze and Green Classify High Detail tools:

  • The Maximum Epochs expert mode training parameter is introduced.

  • The available value range for the Patience Epochs expert mode training parameter is changed to 0–100.

  • The Epoch Count and the Minimum Epochs training parameters are deprecated.

• Resizing option for Red Analyze High Detail tools

  This release introduces the Resizing tool parameters to the Red Analyze High Detail tool to allow for resizing large images. With the Resizing parameters, VisionPro Deep Learning downsizes the image, the marking, and label sizes to make the training and processing time faster. Using smaller image sizes can decrease the accuracy of the tool.

  Note: The same functionality is also available through the API but it has a different name: adaptive downsampling.

• Import mask

  This release introduces an import mask feature. With this feature, you can apply the same mask file to several views at the same time.

• Hand tool

  This release introduces the hand tool to the Edit Mask and Edit Region editors. With the hand tool, you can pan through an image, allowing for more precise masking and region editing.

Enhancements

• The processing time of Red Analyze High Detail tool is faster compared to previous releases.

• The Small option of the Network Model parameter is improved. The improved small network model size allows for faster processing.

• Resizing options are added to the Parameter Search utility for Green Classify High Detail tool. The resizing options are the same as the Resizing parameters available in the VisionPro Deep Learning application when training a Green Classify High Detail tool.

• The VisionPro Deep Learning Software Help document is updated with a restructured table of contents and revised sections.

Changes

• From the VisionPro Deep Learning 3.3.0 release, all tools must be retrained after each version update, including minor version updates.

• From the VisionPro Deep Learning 3.3.0 release, Cognex provides a documentation installer that is separate from the product installer. To get the latest documentation, you must download and install the documentation installer when it becomes available after each release.

• VisionPro Deep Learning now supports Windows Server 2022.

• The High Detail Quick architecture of the Green Classify tool is deprecated.

• The Downsampling Size processing parameter of the Red Analyze High Detail tool is deprecated.

• The Extra Large option of the Network Model parameter is deprecated for the Red Analyze and Green Classify High Detail tools.

• The Whole Image Processing parameter and patch cropping processing is deprecated. From this release, all tools use whole image processing by default.

• The logic used when resizing images is changed. Now, regardless of the image size ratio, VisionPro Deep Learning sets the ratio based on the maximum image height and width. This change only affects Green Classify High Detail tools.