Known Issues

The following issues have been identified in this release:

 

[GPU]

  • Internal issues on supporting NVIDIA driver versions.
    Due to internal issues related to NVIDIA drivers, all users of VisionPro Deep Learning 2.0 MUST use NVIDIA driver according to the followings:
    • 461.09 or higher version for NVIDIA GeForce® series.
    • R460 U2 (461.09) or higher version for NVIDIA RTX / Quadro® series.
  • C API malfunctioning depending on NVIDIA drivers.
    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.
    • Remove the currently installed NVIDIA driver from your PC.
    • Reboot your PC and install the DCH NVIDIA driver of which version is compatible with VisionPro Deep Learning 2.0 (461.09 or higher, or R460 U2 (461.09) or higher for NVIDIA RTX / Quadro®)
  • 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.

 

[OS]

  • OS compatibility issues on Windows Server 2016
    Due to OS compatibility issues caused from NVIDIA drivers, Windows Server 2016 is not supported on VisionPro Deep Learning 2.0. 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 2.0 servers. Windows Server 2019 is only supported for 'servers' configured to use the Deep Learning Client/Server functionality.

 

[Deep Learning Tools]

  • 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.

  • 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.

  • 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.

  • For Red Analyze High Detail, if the tool crashes with some unknown reason, images in this tool are deleted when the workspace that contains this tool is imported.

  • 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.

  • 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.

  • 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.

  • 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.
  • Once you are done with training High Detail mode of Green Classify Tool, you cannot change the name of the class.
  • 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.
  • 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.
  • 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,
    • feature size=10, sampling density=3, 8192x819 image: 1 vvb file size=94MB
    • feature size=10, sampling density=10, 8192x8192 image: 1 vvb file size=1GB
    • feature size=40, sampling density=3, 8192x8192 image: 1 vvb file size=23KB

 

[API]

  • Several minor issues with using APIs. See FAQ for the details.

  • Running Example.Runtime.Remote.Console returns results to only 4 decimal places, not 6 as in the previous VisionPro Deep Learning release.

 

[Workspace]

  • If the tool is not saved well because of unspecific reasons, you cannot export or clone the workspace/tool.
    You can avoid this problem by processing the tools again and save the workspace.
  • Boost::filesystem::rename error.
    This 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.
  • Group is shown in the workspace imported from ViDi Suite.

    • If you import workspaces including "Group" feature which are made 4.1 and before, "Group" is still displayed in Cognex Deep Learning Studio 2.0, but you cannot modify this.

    • See the FAQ for the workaround for this issue.

    • It is recommended to delete "Group" in the previous workspaces and import to Cognex Deep Learning Studio 2.0.

  • Especially with bigger than 10GB workspaces,
    • Cloning work does not end when you clone tools in different big workspaces at the same time.
    • Some images are not included in the created report.

 

[Report]

  • 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.

[Migration]

  • 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 Learning2.0.

 

[Integration]

  • 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.

 

[VisionPro Deep Learning Service]

  • 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.

  • 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.

  • 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.