README.md

M11 DEMO #3

• Supervised Form Recognizer

This code is provided for demo purposes only for course AI-102.

Requirements

• Azure Subscription
• Python

Parsing and Train model.

1. Create a storage account and upload files from the invoice folder to the new container. Create a SAS for the created container.

2. Create Form Recognizer resource and retrieve key and endpoint values

3. Login in Form Recognizer sample labeling tool

4. Click on connected settings and create connection to storage account with SAS token generated before. Details provided in the tutorial

5. Create a new custom project to train models with labels. Provide key and endpoint you copied early

6. Click on 'Tags editor' and from apply labels to the text the details in tutorial

1. You need to have 5 documents labeled to train the model. Complete training for model as explained in tutorial

1. Finally analyze trained form by uploading test document test_invoice.pdf as explained in tutorial

Note that the recognizer can parse and retrieve labels from table's rows with specific name. eg. backup cost.

Test from client (Python)

1. From the labeling tool copy Model ID from model completed above. The Model ID is located on Train page.

2. From Azure portal copy endpoint and key for Form Recognizer created above.

3. Update Python script with copied values as following:

   # Endpoint URL
   endpoint = ""
   # Subscription Key
   apim_key = ""
    # Model ID
   model_id = ""

1. Run the script by following command from terminal:

   py .\analyze-81e0.py .\test_invoice.pdf -o result.json

1. Observe generated field results.json Each filed as PO Number and Address should present on the top of the document with parsed values.