Azure Form Recognizer Table train

Question

Hi All,

I am creating custom region mapping train model using form recognizer studio. I would like to train dynamic tables situated in either one or two pages. kindly suggest best suitable approach to train and consume in custom model training.

Thank you.

Answer 1

Hello @PRASHANTHI THUMMA

Thanks for reaching out to us. As for your question about training dynamic tables situated in either one or two pages, Form Recognizer is designed to handle tables of varying sizes and layouts. However, it is important to ensure that your training data includes examples of the types of tables you expect to encounter in your documents. You may need to experiment with different training parameters to achieve the best results for your specific use case.

To train a custom model for table extraction using Form Recognizer, you can follow these steps:

Collect training data: Gather a set of labeled training data that includes examples of the tables you want to extract. You can use the Form Recognizer labeling tool to label the data.

Create a new custom model: In the Form Recognizer studio, create a new custom model and upload your labeled training data.

Train the model: Start the training process for your custom model. The training process will analyze your labeled data and create a model that can recognize tables in new documents.

Test and refine the model: After the training process is complete, test your model on new documents to see how well it performs. If necessary, you can refine the model by adding more training data or adjusting the training parameters.

Once you have trained your custom model, you can use the Form Recognizer API to extract tables from new documents. The API provides a variety of options for specifying the input document and controlling the output format of the extracted tables.

Reference - https://learn.microsoft.com/en-us/azure/applied-ai-services/form-recognizer/how-to-guides/build-a-custom-model?view=form-recog-3.0.0

I hope this helps.

Regards,

Yutong

-Please kindly accept the answer if you feel helpful to support the community, thanks a lot.