Install and run containers

The following code sample is a self-contained docker compose example to run the Document Intelligence Layout container. With docker compose, you use a YAML file to configure your application's services. Then, with the docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your Layout container instance.

version: "3.9"
services:
  azure-form-recognizer-read:
    container_name: azure-form-recognizer-read
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/read-3.1
    environment:
      - EULA=accept
      - billing={FORM_RECOGNIZER_ENDPOINT_URI}
      - apiKey={FORM_RECOGNIZER_KEY}
    ports:
      - "5000:5000"
    networks:
      - ocrvnet
networks:
  ocrvnet:
    driver: bridge

Now, you can start the service with the docker compose command:

docker-compose up

The following code sample is a self-contained docker compose example to run the Document Intelligence General Document container. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your General Document and Layout container instances.

version: "3.9"
services:
  azure-cognitive-service-document:
    container_name: azure-cognitive-service-document
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/document-3.0
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}
        - AzureCognitiveServiceLayoutHost=http://azure-cognitive-service-layout:5000
    ports:
      - "5000:5050"
  azure-cognitive-service-layout:
     container_name: azure-cognitive-service-layout
     image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/layout-3.0
     environment:
         - EULA=accept
         - billing={FORM_RECOGNIZER_ENDPOINT_URI}
         - apiKey={FORM_RECOGNIZER_KEY}

Now, you can start the service with the docker compose command:

docker-compose up

Given the resources on the machine, the General Document container might take some time to start up.

The following code sample is a self-contained docker compose example to run the Document Intelligence Layout container. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your Layout container instance.

version: "3.9"
services:
  azure-form-recognizer-layout:
    container_name: azure-form-recognizer-layout
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/layout-3.1
    environment:
      - EULA=accept
      - billing={FORM_RECOGNIZER_ENDPOINT_URI}
      - apiKey={FORM_RECOGNIZER_KEY}
    ports:
      - "5000:5000"
    networks:
      - ocrvnet
networks:
  ocrvnet:
    driver: bridge

Now, you can start the service with the docker compose command:

docker-compose up

The following code sample is a self-contained docker compose example to run the Document Intelligence Invoice container. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your Invoice and Layout container instances.

You must use 3.1 GA Layout image as an upstream for both 3.0 GA and 3.1 GA Invoice models.

version: "3.9"
services:
  azure-cognitive-service-invoice:
    container_name: azure-cognitive-service-invoice
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/invoice-3.1
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}
        - AzureCognitiveServiceLayoutHost=http://azure-cognitive-service-layout:5000
    ports:
      - "5000:5050"
  azure-cognitive-service-layout:
    container_name: azure-cognitive-service-layout
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/layout-3.1
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}

Now, you can start the service with the docker compose command:

docker-compose up

The following code sample is a self-contained docker compose example to run the Document Intelligence General Document container. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your Receipt and Read container instances.

You can use 3.1 GA Layout image as an upstream instead of Read image.

version: "3.9"
services:
  azure-cognitive-service-receipt:
    container_name: azure-cognitive-service-receipt
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/receipt-3.1
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}
        - AzureCognitiveServiceReadHost=http://azure-cognitive-service-read:5000
    ports:
          - "5000:5050"
    azure-cognitive-service-read:
      container_name: azure-cognitive-service-read
      image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/read-3.1
      environment:
          - EULA=accept
          - billing={FORM_RECOGNIZER_ENDPOINT_URI}
          - apiKey={FORM_RECOGNIZER_KEY}

Now, you can start the service with the docker compose command:

docker-compose up

The following code sample is a self-contained docker compose example to run the Document Intelligence General Document container. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration. Enter {FORM_RECOGNIZER_ENDPOINT_URI} and {FORM_RECOGNIZER_KEY} values for your ID and Read container instances.

You can use 3.1 GA Layout image as an upstream instead of Read image.

version: "3.9"
services:
  azure-cognitive-service-id-document:
      container_name: azure-cognitive-service-id-document
      image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/id-document-3.1
      environment:
          - EULA=accept
          - billing={FORM_RECOGNIZER_ENDPOINT_URI}
          - apiKey={FORM_RECOGNIZER_KEY}
          - AzureCognitiveServiceReadHost=http://azure-cognitive-service-read:5000
      ports:
          - "5000:5050"
  azure-cognitive-service-read:
      container_name: azure-cognitive-service-read
      image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/read-3.1
      environment:
          - EULA=accept
          - billing={FORM_RECOGNIZER_ENDPOINT_URI}
          - apiKey={FORM_RECOGNIZER_KEY}

Now, you can start the service with the docker compose command:

docker-compose up

version: "3.9"
services:
  azure-cognitive-service-invoice:
    container_name: azure-cognitive-service-businesscard
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/businesscard-3.0
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}
        - AzureCognitiveServiceLayoutHost=http://azure-cognitive-service-layout:5000
    ports:
      - "5000:5050"
  azure-cognitive-service-layout:
    container_name: azure-cognitive-service-layout
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/layout-3.0
    environment:
        - EULA=accept
        - billing={FORM_RECOGNIZER_ENDPOINT_URI}
        - apiKey={FORM_RECOGNIZER_KEY}

In addition to the prerequisites, you need to do the following to process a custom document:

Create a folder and store the following files

• .env
• nginx.conf
• docker-compose.yml

Create a folder and store your input data

• Name this folder files.
• We reference the file path for this folder as {FILE_MOUNT_PATH}.
• Copy the file path in a convenient location, you need to add it to your .env file. As an example if the folder is called files, located in the same folder as the docker-compose file, the .env file entry is FILE_MOUNT_PATH="./files"

Create a folder to store the logs written by the Document Intelligence service on your local machine

• Name this folder output.
• We reference the file path for this folder as {OUTPUT_MOUNT_PATH}.
• Copy the file path in a convenient location, you need to add it to your .env file. As an example if the folder is called output, located in the same folder as the docker-compose file, the .env file entry is OUTPUT_MOUNT_PATH="./output"

Create a folder for storing internal processing shared between the containers

• Name this folder shared.
• We reference the file path for this folder as {SHARED_MOUNT_PATH}.
• Copy the file path in a convenient location, you need to add it to your .env file. As an example if the folder is called shared, located in the same folder as the docker-compose file, the .env file entry is SHARED_MOUNT_PATH="./shared"

Create a folder for the Studio to store project related information

• Name this folder db.
• We reference the file path for this folder as {DB_MOUNT_PATH}.
• Copy the file path in a convenient location, you need to add it to your .env file. As an example if the folder is called db, located in the same folder as the docker-compose file, the .env file entry is DB_MOUNT_PATH="./db"

Create an environment file

1. Name this file .env.

2. Declare the following environment variables:

SHARED_MOUNT_PATH="./shared"
OUTPUT_MOUNT_PATH="./output"
FILE_MOUNT_PATH="./files"
DB_MOUNT_PATH="./db"
FORM_RECOGNIZER_ENDPOINT_URI="YourFormRecognizerEndpoint"
FORM_RECOGNIZER_KEY="YourFormRecognizerKey"
NGINX_CONF_FILE="./nginx.conf"

Create an nginx file

1. Name this file nginx.conf.

2. Enter the following configuration:

worker_processes 1;

events { worker_connections 1024; }

http {

    sendfile on;
    client_max_body_size 90M;
    upstream docker-custom {
        server azure-cognitive-service-custom-template:5000;
    }

    upstream docker-layout {
        server  azure-cognitive-service-layout:5000;
    }

    server {
        listen 5000;

        location = / {
            proxy_set_header Host $host:$server_port;
            proxy_set_header Referer $scheme://$host:$server_port;
            proxy_pass http://docker-custom/;
        }

        location /status {
            proxy_pass http://docker-custom/status;
        }

        location /test {
            return 200 $scheme://$host:$server_port;
        }

        location /ready {
            proxy_pass http://docker-custom/ready;
        }

        location /swagger {
            proxy_pass http://docker-custom/swagger;
        }

        location /formrecognizer/documentModels/prebuilt-layout {
            proxy_set_header Host $host:$server_port;
            proxy_set_header Referer $scheme://$host:$server_port;

            add_header 'Access-Control-Allow-Origin' '*' always;
            add_header 'Access-Control-Allow-Headers' 'cache-control,content-type,ocp-apim-subscription-key,x-ms-useragent' always;
            add_header 'Access-Control-Allow-Methods' 'GET, POST, OPTIONS' always;
            add_header 'Access-Control-Expose-Headers' '*' always;

            if ($request_method = 'OPTIONS') {
                return 200;
            }

            proxy_pass http://docker-layout/formrecognizer/documentModels/prebuilt-layout;
        }

        location /formrecognizer/documentModels {
            proxy_set_header Host $host:$server_port;
            proxy_set_header Referer $scheme://$host:$server_port;

            add_header 'Access-Control-Allow-Origin' '*' always;
            add_header 'Access-Control-Allow-Headers' 'cache-control,content-type,ocp-apim-subscription-key,x-ms-useragent' always;
            add_header 'Access-Control-Allow-Methods' 'GET, POST, OPTIONS, DELETE' always;
            add_header 'Access-Control-Expose-Headers' '*' always;

            if ($request_method = 'OPTIONS') {
                return 200;
            }

            proxy_pass http://docker-custom/formrecognizer/documentModels;
        }

        location /formrecognizer/operations {
            add_header 'Access-Control-Allow-Origin' '*' always;
            add_header 'Access-Control-Allow-Headers' 'cache-control,content-type,ocp-apim-subscription-key,x-ms-useragent' always;
            add_header 'Access-Control-Allow-Methods' 'GET, POST, OPTIONS, PUT, DELETE, PATCH' always;
            add_header 'Access-Control-Expose-Headers' '*' always;

            if ($request_method = OPTIONS ) {
                return 200;
            }

            proxy_pass http://docker-custom/formrecognizer/operations;
        }
    }
}

Create a docker compose file

1. Name this file docker-compose.yml

2. The following code sample is a self-contained docker compose example to run Document Intelligence Layout, Studio, and Custom template containers together. With docker compose, you use a YAML file to configure your application's services. Then, with docker-compose up command, you create and start all the services from your configuration.

version: '3.3'
services:
  nginx:
    image: nginx:alpine
    container_name: reverseproxy
    volumes:
      - ${NGINX_CONF_FILE}:/etc/nginx/nginx.conf
    ports:
      - "5000:5000"
  layout:
    container_name: azure-cognitive-service-layout
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/layout-3.0:latest
    environment:
      eula: accept
      apikey: ${FORM_RECOGNIZER_KEY}
      billing: ${FORM_RECOGNIZER_ENDPOINT_URI}
      Logging:Console:LogLevel:Default: Information
      SharedRootFolder: /shared
      Mounts:Shared: /shared
      Mounts:Output: /logs
    volumes:
      - type: bind
        source: ${SHARED_MOUNT_PATH}
        target: /shared
      - type: bind
        source: ${OUTPUT_MOUNT_PATH}
        target: /logs
    expose:
      - "5000"

  custom-template:
    container_name: azure-cognitive-service-custom-template
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/custom-template-3.0:latest
    restart: always
    depends_on:
      - layout
    environment:
      AzureCognitiveServiceLayoutHost: http://azure-cognitive-service-layout:5000
      eula: accept
      apikey: ${FORM_RECOGNIZER_KEY}
      billing: ${FORM_RECOGNIZER_ENDPOINT_URI}
      Logging:Console:LogLevel:Default: Information
      SharedRootFolder: /shared
      Mounts:Shared: /shared
      Mounts:Output: /logs
    volumes:
      - type: bind
        source: ${SHARED_MOUNT_PATH}
        target: /shared
      - type: bind
        source: ${OUTPUT_MOUNT_PATH}
        target: /logs
    expose:
      - "5000"

  studio:
    container_name: form-recognizer-studio
    image: mcr.microsoft.com/azure-cognitive-services/form-recognizer/studio:3.0
    environment:
      ONPREM_LOCALFILE_BASEPATH: /onprem_folder
      STORAGE_DATABASE_CONNECTION_STRING: /onprem_db/Application.db
    volumes:
      - type: bind
        source: ${FILE_MOUNT_PATH} # path to your local folder
        target: /onprem_folder
      - type: bind
        source: ${DB_MOUNT_PATH} # path to your local folder
        target: /onprem_db
    ports:
      - "5001:5001"
    user: "1000:1000" # echo $(id -u):$(id -g)

The custom template container and Layout container can use Azure Storage queues or in memory queues. The Storage:ObjectStore:AzureBlob:ConnectionString and queue:azure:connectionstring environment variables only need to be set if you're using Azure Storage queues. When running locally, delete these variables.

Ensure the service is running

To ensure that the service is up and running. Run these commands in an Ubuntu shell.

$cd <folder containing the docker-compose file>

$source .env

$docker-compose up

Custom template containers require a few different configurations and support other optional configurations.

Use the Document Intelligence Studio to train a model

• Gather a set of at least five forms of the same type. You use this data to train the model and test a form. You can use a sample data set (download and extract sample_data.zip).

• Once you can confirm that the containers are running, open a browser and navigate to the endpoint where you have the containers deployed. If this deployment is your local machine, the endpoint is [http://localhost:5001](http://localhost:5001).

• Select the custom extraction model tile.

• Select the Create project option.

• Provide a project name and optionally a description

• On the "configure your resource" step, provide the endpoint to your custom template model. If you deployed the containers on your local machine, use this URL [http://localhost:5000](http://localhost:5000).

• Provide a subfolder for where your training data is located within the files folder.

• Finally, create the project

You should now have a project created, ready for labeling. Upload your training data and get started labeling. If you're new to labeling, see build and train a custom model.

Using the API to train

If you plan to call the APIs directly to train a model, the custom template model train API requires a base64 encoded zip file that is the contents of your labeling project. You can omit the PDF or image files and submit only the JSON files.

Once you have your dataset labeled and *.ocr.json, *.labels.json and fields.json files added to a zip, use the PowerShell commands to generate the base64 encoded string.

$bytes = [System.IO.File]::ReadAllBytes("<your_zip_file>.zip")
$b64String = [System.Convert]::ToBase64String($bytes, [System.Base64FormattingOptions]::None)

Use the build model API to post the request.

POST http://localhost:5000/formrecognizer/documentModels:build?api-version=2023-07-31

{
    "modelId": "mymodel",
    "description": "test model",
    "buildMode": "template",

    "base64Source": "<Your base64 encoded string>",
    "tags": {
       "additionalProp1": "string",
       "additionalProp2": "string",
       "additionalProp3": "string"
     }
}