How to run the Azure Computer Vision engine On-Premise

The Azure Computer Vision engine, which MetaServer uses for the Extract Text (Azure Computer Vision) rule, can be run on-premise using Containers through the Docker engine.

Running the engine on-premise can be useful for security and data governance requirements.

1) An active Azure Computer Vision resource: you will need to have a valid key for an Azure Computer Vision resource. If you don’t already applied for one, you can find instructions on our Extract Text (Azure Computer Vision) online help page.

Once you have your key, please keep the following information on hand:

{API_KEY}: One of the two available resource keys on the Keys and Endpoint page.

{ENDPOINT_URI}: The endpoint as provided on the Keys and Endpoint page.

2) The Docker Engine on your host system: you need the Docker Engine installed on your host computer.

You can find instructions on how to install the Docker environment through Docker’s online documentation.

NOTE: In case your host computer shuts down and needs to be restarted, by default, you will need to log in and launch the Docker engine manually. To automatically launch the Docker engine at a system (re)start, enable the “Start Docker Desktop when you log in” option in your Docker client’s settings.

3) Approval from Microsoft to run the Azure Cognitive Services container on your system: you will need to apply for permission to run the container from Microsoft through the following online request form.

The form requests information about you, your company, and the user scenario for which you’ll use the container. After you submit the form, the Microsoft Azure Cognitive Services team will review it and email you with a decision within 10 business days.

After you’re approved, you will be able to run the container after downloading it from the Microsoft Container Registry (MCR). This will be described in more detail in the next step.

You won’t be able to run the container if your Azure subscription has not been approved.

After you’ve received permission from Microsoft to run the ACV container on your system, you can install and run the container through your Windows Command Prompt.

IMPORTANT: make sure the logged-in user has the appropriate installation rights in order to proceed. If you don’t have permission, please contact your IT administrator to (temporarily) elevate your user rights.

If you’re not sure where to find this information, please refer to the following online help.

Step 1: Enter one of the following commands in your Windows Command Prompt:

Command 1 (for ACV model version 3.2):

docker run -dit --restart unless-stopped -it -p 5000:5000 --memory 18g --cpus 6 mcr.microsoft.com/azure-cognitive-services/vision/read:3.2 Eula=accept Billing={ENDPOINT_URI} ApiKey={API_KEY}

This command:

• Downloads and runs the Read OCR container from the ACV 3.2 model’s container image.
• Allocates 6 CPU core and 18 gigabytes (GB) of memory.
• Exposes TCP port 5000 and allocates a pseudo-TTY for the container.
• Automatically restarts the container after it stops (e.g. after a system crash), unless it was manually stopped.

If your system has less or more than 6 CPU cores or 18 GB of memory, please adjust this number for the corresponding tags (--memory 18g and --cpus 6) accordingly.

As an example, we have allocated TCP port 5000, but you can change it to any available port. To change this port, please adjust this number for the corresponding tag (-p 5000:5000) accordingly.

Command 2 (for ACV preview model version 3.2 2021-09-30):

docker run -dit --restart unless-stopped -it -p 5000:5000 --memory 18g --cpus 6 mcr.microsoft.com/azure-cognitive-services/vision/read:3.2-model-2021-09-30-preview Eula=accept Billing={ENDPOINT_URI} ApiKey={API_KEY}

This command:

• Downloads and runs the Read OCR container from the ACV preview model version 3.2 2021-09-30’s container image.
• Allocates 6 CPU core and 18 gigabytes (GB) of memory.
• Exposes TCP port 5000 and allocates a pseudo-TTY for the container.
• Automatically restarts the container after it stops (e.g. after a system crash), unless it was manually stopped.

If your system has less or more than 6 CPU cores or 18 GB of memory, please adjust this number for the corresponding tags (--memory 18g and --cpus 6) accordingly.

As an example, we have allocated TCP port 5000, but you can change it to any available port. To change this port, please adjust this number for the corresponding tag (-p 5000:5000) accordingly.

After you run these commands, wait until it returns the confirmation message “Application started”:

Step 2: you can choose to close the Command Prompt window or leave it open. The container will keep running either way.

Step 3: you can always check, stop, restart or delete your running containers in your Docker desktop client under the “Container / Apps” tab.

Step 4: as a final test, run your local URL in a browser. In our example, we allocated TCP port 5000, so we use the URL “http:\\localhost:5000”.

If everything went correctly, it should show you the following splash page:

Step 3: Select the model you’ve used for your local ACV engine. In our example, we’ve used the “3.2 2021-09-30-preview” model, so the settings for this are:

Model: Preview
Preview version: 2021-09-30-preview

If you use the general available model, you just need to select “General Available” as your Model. MetaServer automatically uses the latest model version, so you don’t need to specify it.

If your server needs to be fully disconnected from the internet, disconnected containers will enable you to use Azure services completely offline.

You can find more information about this capability and how to request access to a disconnected container here:
https://learn.microsoft.com/en-us/azure/cognitive-services/containers/disconnected-containers