How to use WEDA Edge AI Containers

Prepare a Ready-to-use application from Advantech Container Catalog (ACC) for deployment with WEDA. This guide uses Intel OpenVINO Retail AI Self-Checkout as an example and finishes with the application image, Compose file, host files and environment settings needed for deployment.

Choose Ready-to-use when you want to deploy an existing application. A Ready-to-develop project must be built before deployment; a Partner container follows its provider's trial or solution instructions. See What is Advantech Container Catalog? for the resource types and examples.

Browse and Select a Container

1. Open Advantech Container Catalog.
2. Under Container types, select Ready-to-use. Use Clear filters if you need to remove a previous selection.
3. Use the search field, Use cases or hardware filters to narrow the results for your application and target device.
4. Open Intel OpenVINO Powered: Retail AI Self-Checkout from the results.

Select Ready-to-use before choosing an application for direct deployment.

Check the result: The selected filter appears above the results, and the application's detail page provides Overview, Tags and Compatible hardware entries.

Check Hardware and Software Requirements

1. Open Compatible hardware and check your target model. For this example, the list includes MIC-780, SOM-5885 and UBX-330M. Use the live listing when selecting hardware for a particular application version.
2. Return to Overview and read Prerequisites.
3. Compare the application requirements with the software and peripherals available on your target device.

Check the selected application's hardware list rather than assuming compatibility from the chipset vendor alone.

The example's catalog page and repository describe these requirements:

Requirement	What to prepare for this example
Host operating system	Ubuntu 24.04 LTS.
Container tools	Docker Engine and Docker Compose. Git is used to retrieve the example files.
Inference device	Start with CPU mode for initial validation. Intel iGPU or NPU operation requires suitable hardware and the corresponding host drivers.
Display	A target-device desktop and X11 access for the application's preview window and interactive Region of Interest (ROI) selection.
Video input	Prepare a camera if you intend to use live video. The supplied startup script also supports automatic camera detection with a configured fallback video.

Read the selected item's operating system, runtime and accelerator requirements before preparing deployment settings.

Note: This example uses a local display

The application opens a preview window on the target device, where a user draws an ROI and presses Enter to start inference. Prepare the target desktop's display and access settings before using this example. WEDA deployment does not create a desktop session or turn that window into a browser application.

Review Container Configuration

1. Open Tags to review the application's version and download information. If the page displays Sign in to see details, select Login and sign in before continuing.
2. Record the image name and version you intend to use, then compare them with the image reference in the repository's Compose file.
3. Review its environment variables, host paths and device access before preparing the WEDA Stack.

The Tags page requires sign-in to display version and download details.

The Compose file used by this example references this image:

harbor.edgesync.cloud/intel/openvino-retail-self-checkout-ai:2026.4-yolo26n

Use the matching configuration when choosing that version. The following settings come from the same repository snapshot:

Setting	Meaning and preparation
DEVICE	Defaults to CPU. Use the inference device supported by your hardware and installed drivers.
INPUT	Defaults to auto. The startup script attempts camera detection, then selects the fallback video if no camera is found. The repository also documents webcam mode.
FALLBACK_VIDEO	Defaults to automated-small.mp4 for the automatic fallback path. Check the selected image and input files before running.
DISPLAY	Identifies the target desktop's display. Supply the value from the target device's intended desktop session.
/tmp/.X11-unix mount	Provides the host X11 socket used by the preview window. Prepare the corresponding display access on the target.
${HOME}/Downloads:/mnt mount	Maps a host directory into the container. Replace the source with the intended absolute directory on the target device.
./start.sh:/start.sh:ro mount	Provides the startup script. Keep it with the Compose file for local use; for WEDA deployment, copy it to the target and use its absolute host path.
entrypoint	Runs /bin/bash /start.sh. The mounted script is required for this startup configuration.
network_mode and privileged	The supplied configuration uses host networking and privileged access. Review these settings for the target environment.

Check the result: You can identify the selected image version, target display, input mode and host files that the application will use. A path on your workstation is not a substitute for the required file on the target device.

Get the Quick Start and Examples from GitHub

1. On the catalog detail page, open Overview and find Quick Start Guide.
2. Follow Advantech Container Github Repository to the Retail AI Self-Checkout repository.

The catalog's Quick Start Guide links to the application-specific repository and setup files.

3. Retrieve the project files. The checkout below selects the snapshot used for the image and settings described in this guide:

git clone https://github.com/Advantech-Containers/openvino-retail-self-checkout-ai.git
cd openvino-retail-self-checkout-ai
git checkout 56dd1c259ba6e8e39ef33c7c87019e1a886354b0

4. Review these files together before changing the configuration:

File	What to review
README.md	Host prerequisites, quick start, display interaction and the expected ROI selection workflow.
docker-compose.yaml	Image reference, environment variables, mounts and runtime settings.
start.sh	How DEVICE, INPUT and FALLBACK_VIDEO select the inference and input modes.
LICENSE	The repository's license. Also read the README's notes about upstream components, datasets and models.

The repository's local quick start is useful for understanding application behavior. Prepare desktop access according to the target's access-control policy. When deploying through WEDA, use the prepared Compose definition and the WEDA deployment workflow below.

Check the result: Your working directory contains the Compose file and startup script from the same snapshot. Keep any image-version changes aligned with the corresponding application instructions and files.

Prepare for Deployment with WEDA

1. Prepare an Activated and Connected WEDA Node in the organization you will use for deployment. Complete Onboard Your Device if the device is not yet connected.
2. Check that the target can reach the image registry and has the required pull access. Use Registry Management to prepare the registry configuration in that organization.
3. Copy start.sh and any application input files to their intended locations on the target device. In the deployment copy of the Compose file, replace relative and shell-dependent host paths with the corresponding absolute target paths.
4. Prepare the Stack's environment values. For an initial CPU run, use DEVICE=CPU and the intended INPUT mode; provide DISPLAY from the target desktop and any other values required by your selected configuration.
5. Confirm the display session, X11 access and required hardware are available on the target. Keep the application's files and environment values with the deployment configuration you intend to save.
6. Continue with Deploy Your First Application to add or reuse the registry configuration, create the Container Stack and deploy its selected revision. For revision management, use Container Stack Management.

Note: Prepare host files before deployment

Uploading a Compose file creates the Stack definition. It does not transfer files referenced by host bind mounts. For this example, place start.sh on the target before deploying the revision that mounts it.

Check the result: The deployment handoff includes the image and version, prepared Compose file, environment values, registry access and target host files. After deployment, check the selected revision and container status in WEDA, then verify the application's preview and ROI interaction on the target display. Container status alone does not confirm that the vision workflow is operating as intended.