Q&A

Answers to common questions about the AI Manager plugin.

I Don't See Any Bounding Boxes in the Client Application

First make sure that the model actually has something to detect.

For example if you're using a model that is expecting vehicles, nothing will be detected on a camera that is viewing an empty warehouse with no vehicles.

If the camera actually sees something that should be detected you can check if the objects view is active.

If the notifications or another tab is active the bounding boxes will not be displayed. When you switch to the objects tab the bounding boxes should show up.

When a tab other than Objects is selected, bounding boxes do not appear on the video stream

Switch to the Objects tab to see bounding boxes on the video stream

If the Objects tab is not available after freshly enabling the plugin, restart the client first.

If bounding boxes are still missing, work through the plugin checks, system checks, and things to try sections.

The AI Runtime Is Having Trouble Starting: No Matching Architecture Found for Model

This message might appear if you have a model assigned that is not compatible with your current hardware or runtime. The AI Manager normally prevents assigning incompatible models, but this can still happen after moving a device to a different server or changing the server's runtime.

Assign a different compatible model or reselect the runtime to solve this problem.

The AI Runtime Is Having Trouble Starting: AI Manager Failed to Create Listening Socket

The AI Manager tries to create a socket file on the filesystem to communicate with the Network Optix Mediaserver. It will try to create this file at the default location (/tmp/scailable-ipc.sock).

If a file already exists, or the AI Manager lacks permission to write to the default location, this fails.

Making this path available solves the issue. On devices where this is not an option, the file location can be controlled through .ini settings. See the Enable .ini settings or Enable .ini settings instructions.

The AI Runtime Is Having Trouble Starting: If All Else Fails

If the message still appears, there could be a problem with your installation. Try the Linux or Windows manual installation instructions.

Runtime status panel showing the AI Manager has failed to start

How Can I Check If the Metadata Is Generated

I noticed that plugin stops generating metadata for some reason Is there any log or additional information I can collect for you so it would be useful in this case?

For advanced users, it might be advisable to check the output log of the Mediaserver to see if any errors are logged. The log can be gathered by executing:

Linux

sudo journalctl -u networkoptix-metavms-mediaserver.service

This may differ depending on your installed version.

Windows

Contact support for detailed instructions.

Is It Possible to Send Output to a Different Endpoint?

No this is not possible. It is possible to access the data and pass it through. See External Postprocessing

The Bounding Boxes Appear to Be out of Sync with the Stream

This issue occurs when the visualized stream differs from the one used for AI inference. It typically happens when using a test camera. The problem usually arises from a mismatch between the primary and secondary streams, leading to the visualization of a different stream than the one used for AI processing. For instance, if the plugin operates on the secondary stream while you are viewing the primary stream, it can appear as though the bounding boxes are out of sync in the video:

To fix this, either set the plugin to use the Primary stream or view the ‘Low’ resolution stream in the Client:

Select the matching stream in the Nx client to fix bounding box sync

How to Install the Nightly Version?

Only install the nightly version of the AI Manager if the support team requests it. The support team will provide the installation instructions.

Is It Safe to Update?

To update or reinstall the AI Plugin, re-run the install command.

The install script removes the old version of the plugin before installing the new version.

Existing settings carry over as far as possible. Check the settings afterward, because new options may require changes. An example might be a new runtime that is available for your hardware.

Why Are Some Models Not Visible When Assigning to a Device

Models or Model Catalogues must be connected to the Organization that your site is in.

If you have chosen an Organization during upload, and the device you are coming from to select a model is not in any Organization, then only the generally available models will be visible.

If you have chosen an Organization and the Site does not belong to the same Organization you have selected for the model, then the model may not be visible either.

In the upload process you can directly select a Site for the model instead of an Organization. This is not advised, however, because the models will then not be available for the other Sites in your Organization.

To solve this, add the Site to an Organization, and add the uploaded models to a Catalogue connected to that Organization.

I See the Error Message: Failed to Create SHM: Function Not Implemented

The AI Manager fails to start with the error Failed to create SHM: Function not implemented.

This device's kernel configuration omits CONFIG_SYSVIPC=y, which the AI Manager requires for its shared-memory inter-process communication.

Enable CONFIG_SYSVIPC=y in the kernel configuration and recompile the kernel, or use a pre-built OS image that includes it. Ubuntu 20.04+ and Debian have this enabled by default. If you are seeing this error, you are likely on a custom embedded Linux image.