Custom extraction model for ID recognistion too slow

Alejandro Diaz 5

Hello,

We are implementing ID recognition for our national identity card. The model we require, which will be connected via API to our web service, must be highly efficient in performing OCR and field identification.

Initially, we tried using a pre-trained ID model. It works adequately, but does not recognize all the fields.

Next, we decided to use a custom extraction model, which performs perfectly in terms of accuracy. However, it is very slow, taking 10 seconds for evaluation.

Given that we are looking for a fast service and that we need to scan both sides of the identity document, how can we improve the processing speed?

Thank you in advance.

YutongTie-MSFT 51,756 Reputation points

2023-09-23T21:01:21.24+00:00

Hello @Alejandro Diaz

Thanks for reaching out to us, I can understand your scenario, but in case you are not aware of it, there is a model for ID, which should be better than the general model and faster than the custom model, have you tried it? Please take a look the document here.

I hope it helps.

Regards,

Yutong
Alejandro Diaz 5 Reputation points

2023-09-25T06:17:36.28+00:00

Hi @YutongTie-MSFT

thank you very much for your answer. I have tried indeed the model for ID. However, while it is faster, the accuracy is too low to trust its results at a production level.

I am using it for spanish national IDs
YutongTie-MSFT 51,756 Reputation points

2023-10-02T04:33:49.79+00:00

Hello @Alejandro Diaz

Thanks for your response, may I know where are you located and which region you are using?

Regards,

Yutong

Share via

Custom extraction model for ID recognistion too slow

Your answer