You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: articles/applied-ai-services/form-recognizer/concept-custom-neural.md
+14-15Lines changed: 14 additions & 15 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -7,7 +7,7 @@ manager: nitinme
7
7
ms.service: applied-ai-services
8
8
ms.subservice: forms-recognizer
9
9
ms.topic: conceptual
10
-
ms.date: 06/06/2022
10
+
ms.date: 07/11/2022
11
11
ms.author: lajanuar
12
12
ms.custom: references_regions
13
13
recommendations: false
@@ -37,8 +37,8 @@ Custom neural models currently only support key-value pairs and selection marks,
37
37
38
38
With the release of API version **2022-06-30-preview**, custom neural models will support tabular fields (tables):
39
39
40
-
* Models trained with API version 2022-06-30-preview or later will accept tabular field labels.
41
-
* Documents analyzed with custom neural models using API version 2022-06-30-preview or later will produce tabular fields aggregated across the tables.
40
+
* Models trained with API version 2022-06-30-preview or later will accept tabular field labels.
41
+
* Documents analyzed with custom neural models using API version 2022-06-30-preview or later will produce tabular fields aggregated across the tables.
42
42
* The results can be found in the ```analyzeResult``` object's ```documents``` array that is returned following an analysis operation.
43
43
44
44
Tabular fields support **cross page tables** by default:
@@ -50,23 +50,22 @@ Tabular fields are also useful when extracting repeating information within a do
50
50
51
51
## Supported regions
52
52
53
-
Starting on 1 August 2022, Form Recognizer custom neural model training will only be available in the following Azure regions until further notice:
54
-
55
-
* Brazil South
56
-
* Canada Central
57
-
* Central India
58
-
* Japan East
59
-
* North Europe
60
-
* South Central US
61
-
* Southeast Asia
53
+
As of August 01 2022, Form Recognizer custom neural model training will only be available in the following Azure regions until further notice:
62
54
55
+
* Brazil South
56
+
* Canada Central
57
+
* Central India
58
+
* Japan East
59
+
* North Europe
60
+
* South Central US
61
+
* Southeast Asia
63
62
64
63
> [!TIP]
65
64
> You can [copy a model](disaster-recovery.md) trained in one of the select regions listed above to **any other region** and use it accordingly.
66
65
67
66
## Best practices
68
67
69
-
Custom neural models differ from custom template models in a few different ways. The custom template or model relies on a consistent visual template to extract the labeled data. Custom neural models support structured, semi-structured, and unstructured documents to extract fields. When you're choosing between the two model types, start with a neural model and test to determine if it supports your functional needs.
68
+
Custom neural models differ from custom template models in a few different ways. The custom template or model relies on a consistent visual template to extract the labeled data. Custom neural models support structured, semi-structured, and unstructured documents to extract fields. When you're choosing between the two model types, start with a neural model, and test to determine if it supports your functional needs.
70
69
71
70
### Dealing with variations
72
71
@@ -87,7 +86,6 @@ Value tokens/words of one field must be either
87
86
88
87
Values in training cases should be diverse and representative. For example, if a field is named "date", values for this field should be a date. synthetic value like a random string can affect model performance.
89
88
90
-
91
89
## Current Limitations
92
90
93
91
* The model doesn't recognize values split across page boundaries.
* The [Visual Studio IDE](https://visualstudio.microsoft.com/vs/) or current version of [.NET Core](https://dotnet.microsoft.com/download/dotnet-core).
29
29
* An Azure Storage blob that contains a set of training data. See [Build a training data set for a custom model](../../build-training-data-set.md) for tips and options for putting together your training data set. For this project, you can use the files under the **Train** folder of the [sample data set](https://go.microsoft.com/fwlink/?linkid=2090451) (download and extract *sample_data.zip*).
30
30
* Once you have your Azure subscription, <ahref="https://portal.azure.com/#create/Microsoft.CognitiveServicesFormRecognizer"title="Create a Form Recognizer resource"target="_blank">create a Form Recognizer resource </a> in the Azure portal to get your key and endpoint. After it deploys, select **Go to resource**.
31
-
* You will need the key and endpoint from the resource you create to connect your application to the Form Recognizer API. Paste your key and endpoint into the code below later in the project.
31
+
* You'll need the key and endpoint from the resource you create to connect your application to the Form Recognizer API. Paste your key and endpoint into the code below later in the project.
32
32
* You can use the free pricing tier (`F0`) to try the service, and upgrade later to a paid tier for production.
33
33
> [!TIP]
34
-
> Create a Cognitive Services resource if you plan to access multiple cognitive services under a single endpoint/key. For Form Recognizer access only, create a Form Recognizer resource. Please note that you'll need a single-service resource if you intend to use [Azure Active Directory authentication](../../../active-directory/authentication/overview-authentication.md).
34
+
> Create a Cognitive Services resource if you plan to access multiple cognitive services under a single endpoint/key. For Form Recognizer access only, create a Form Recognizer resource. Please note that you'll need a single-service resource if you intend to use [Azure Active Directory authentication](../../../../active-directory/authentication/overview-authentication.md).
35
35
36
36
## Setting up
37
37
@@ -78,7 +78,7 @@ In the application's **Program** class, create variables for your resource's key
@@ -88,11 +88,11 @@ With Form Recognizer, you can create two different client types. The first, `For
88
88
89
89
### FormRecognizerClient
90
90
91
-
`FormRecognizerClient` provides operations for:
91
+
`FormRecognizerClient` provides the following operations:
92
92
93
-
*Recognizing form fields and content, using custom models trained to analyze your custom forms. These values are returned in a collection of `RecognizedForm` objects. See example [Analyze custom forms](#analyze-forms-with-a-custom-model).
94
-
*Recognizing form content, including tables, lines and words, without the need to train a model. Form content is returned in a collection of `FormPage` objects. See example [Analyze layout](#analyze-layout).
95
-
*Recognizing common fields from US receipts, business cards, invoices, and ID documents using a pre-trained model on the Form Recognizer service.
93
+
*Recognize form fields and content, using custom models trained to analyze your custom forms. These values are returned in a collection of `RecognizedForm` objects. See example [Analyze custom forms](#analyze-forms-with-a-custom-model).
94
+
*Recognize form content, including tables, lines and words, without the need to train a model. Form content is returned in a collection of `FormPage` objects. See example [Analyze layout](#analyze-layout).
95
+
*Recognize common fields from US receipts, business cards, invoices, and ID documents using a pre-trained model on the Form Recognizer service.
96
96
97
97
### FormTrainingClient
98
98
@@ -291,7 +291,7 @@ The following code processes the ID document at the given URI and prints the maj
291
291
292
292
## Train a custom model
293
293
294
-
This section demonstrates how to train a model with your own data. A trained model can output structured data that includes the key/value relationships in the original form document. After you train the model, you can test and retrain it and eventually use it to reliably extract data from more forms according to your needs.
294
+
This section demonstrates how to train a model with your own data. A trained model can output structured data that includes the key/value relationships in the original form document. After you train the model, you can test, retrain, and eventually use it to reliably extract data from more forms according to your needs.
295
295
296
296
> [!NOTE]
297
297
> You can also train models with a graphical user interface such as the [Form Recognizer Sample Labeling tool](../../label-tool.md).
@@ -368,7 +368,7 @@ Submodel Form Type: form-0
368
368
369
369
### Train a model with labels
370
370
371
-
You can also train custom models by manually labeling the training documents. Training with labels leads to better performance in some scenarios. To train with labels, you need to have special label information files (`\<filename\>.pdf.labels.json`) in your blob storage container alongside the training documents. The [Form Recognizer Sample Labeling tool](../../label-tool.md) provides a UI to help you create these label files. Once you have them, you can call the `StartTrainingAsync` method with the `uselabels` parameter set to `true`.
371
+
You can also train custom models by manually labeling the training documents. Training with labels leads to better performance in some scenarios. To train with labels, you need to have special label information files (`\<filename\>.pdf.labels.json`) in your blob storage container alongside the training documents. The [Form Recognizer Sample Labeling tool](../../label-tool.md) provides a UI to help you create these label files. Once you've them, you can call the `StartTrainingAsync` method with the `uselabels` parameter set to `true`.
If you want to clean up and remove a Cognitive Services subscription, you can delete the resource or resource group. Deleting the resource group also deletes any other resources associated with it.
598
+
If you want to cleanup and remove a Cognitive Services subscription, you can delete the resource or resource group. Deleting the resource group also deletes any other resources associated with it.
0 commit comments