What's New In The Form Recognizer Sample Labelling Tool v2.1?

Introduction 

 

Let's go through the new version v2.1 of the sample labeling tool and look at the updated features. The web app version of Sample Labelling Tool v2.1 in this link, (click here), can be accessed for testing, evaluation, or academic purpose.

 

The improvements in version 2.1 which are currently in the preview include:

1. Compose Trained Models.
2. Giving Models a Model Name to identify models.
3. Identify whether the selection mark is "selected" or "unselected".
4. Option for Sharing Project with string.
   
   The landing page of version 2.1 of the sample labeling tool is shown in the below screenshot:
   
   .

Compose Trained Models

 

This can be mentioned as the major change while moving from version v2.0 to 2.1. Let's get deeper into it.

 

In version 2.0 we can train and test models but each model will be having a separate Model ID and thus the headache to select the model corresponding to the input was still in the hands of a developer. i.e; I have trained models for invoices from an "xyz" company and another "abc" company. These two still have a separate Model ID, and the developer needs to use some classification algorithm to figure out whether the inputted invoice is from the "xyz" company or another "abc" company. Then, we call the form recognizer with the corresponding Model ID.

 

But in version 2.1, we can compose multiple models and get a single Model ID which we can use while integrating without an application. Now the headache of classifying input is taken care of by the Form Recognizer. i.e; in the above scenario we get a single Model ID for two models after composing.

 

How to Compose Trained Models

1. First, train the models separately as different projects. Consider we have trained two models with names "Statements" and "Project Statement".
2. Click the "Compose" button between the "Train" button and the "Analyze" button. There you can see all the models listed. Here is a screenshot of the same:
   
   
   
   
3. Now select the models you need to compose. Left of the Model Id column in the table there will be a tick mark if got selected.
4. Select "Compose" on the top left of the center tab above the table and left of the "Refresh" button.
5. Now a prompt to enter the new model name appears. Enter a new name and double-check if the models to compose are listed in the prompt are correct.
6. Now Click "Compose" on the prompt.
7. After composing is completed, the model will be added to the model list. Please click "Refresh" on top of the table if you don't find the same.
8. The composed model will be having a symbol on the left of the model name indicating it is a composed model. You can see the last model named "8620 + CC" in the above screenshot is a composed model.
9. You can copy the Model ID from here and use it in the integration of the application developed. If you use Mozilla Firefox you will find it difficult to copy so use some other browsers (I use Google Chrome).
10. Now you can test the same by selecting this model from from "Analyze" page (by clicking the "Analyze" icon on the left pane below the "Compose" button) of the "Sample Form Recognition Tool".

Giving Names to Models

 

In version 2.0, we were identifying models with Model ID. As models and projects are related, we didn't have an issue in identifying models, as it does not have scope out of the project. We can easily identify it with Project Name in version 2.0 as the model doesn't have scope outside of the project.

 

In version 2.1 models are having scope out of the project with the introduction of "Compose" so now it's necessary to identify the model with some name other than Model ID (Model ID is auto-generated thus it's not easy to identify by humans).

 

We now identify models with names and are asked while we train a model. Even though it's not a compulsory field, it's recommended to use the same as it makes our task easy.

 

Identify Selection Marks

 

Some invoices/documents are having selection marks like radio buttons or checkboxes. The selected and unselected status of these selection marks is some valuable information. Ex: We usually have a checkbox in an application form to collect data related to "Gender" or "Marital Status". In order to get such information using Form Recognizer, we can use this feature.

 

Share Project

 

You can share a project by just clicking the share button on the top right of the home page. The same is highlighted with the red square in the below screenshot. which will copy the string to the clipboard On the machine/browser we need to import the project/model. Click on "Open Cloud Project" in the homepage and paste the string which is already on the clipboard. Click the "OK" button to start the import.

 

Go to the official documentation for more details.

 

I hope you loved this article. Please comment if you faced any difficulties while following along. I'm also waiting for your valuable feedback.