Is it possible to increase form recognizer performance (prebuilt-invoice model) ?

Question

I am using the python sdk for form recognizer to extract datas such as total, invoiceId etc ... from invoices with the prebuilt invoice model.

It works perfectly but I'm experiencing a very slow performance: 10s to 20s for a one page document ( 24kb image or pdf ).

Is it the normal duration of document scanning ? If not, is there any solution to increase that performance ?

For now, I am using the free pricing tier in West Europe region.

Thanks

Answer 1

The duration of document scanning can depend on several factors, such as the size and complexity of the document, the network latency, and the performance of the processing resources being used. However, a 10-20 second processing time for a single page document does seem on the higher side.

Posting possible ways to improve the performance of your form recognizer process:

• Upgrade to a higher pricing tier with more processing power and resources. The free pricing tier is limited in terms of processing power and can result in slower processing times.
• Use a different region that may have better processing capabilities or lower latency.
• Optimize your code and usage of the Form Recognizer API to minimize the processing time. For example, you can try using batch processing to process multiple documents in parallel.
• Optimize the input documents by reducing their size or complexity. For example, you can try converting the documents to black and white or reducing the resolution.
• Cache the results of the processing so that you don't have to process the same document again if it's already been processed.
• Consider using other OCR libraries or services that may have faster processing times.