What is required to train a custom model using Form Recognizer?

To train a custom model using Azure Form Recognizer, it is necessary to provide labeled examples in the form of JSON files that correspond to the form documents you wish to analyze. These JSON files contain annotations that give context to the unstructured data present in the forms, outlining the areas where specific information can be found.

Creating a custom model without these labeled JSON files would mean that there is no clear guiding information for the model to learn from, which is essential for it to understand the structure and capture the relevant data accurately. This training process allows the model to adapt to the specific layouts and content of the forms that it will later recognize, enabling it to extract the required information effectively.

In the context of the other options provided, training a model without additional resources like those JSON files would not yield meaningful results, nor would simply utilizing SDKs or template models be sufficient without this crucial labeled data.