Nanonets_OCR on 2 * A100 GPU

vision
1

Hi everybody,

I am using Nanonets_ocr from huggingface to perform some ocr task. The model is quite good at extracting data but the problem is that, it is taking too long to extract the data (almost 2 minutes).

This is the application set up:

The model is running on 2* A100 nvidia GPU server, and I have exposed a flask api where user can upload one image and call the api, in the back end the model extract specific data and return it to the front end developper.

Can somebody helps me to achieve better accuracy ?

2

Is the issue the long extraction time or the model accuracy? In the former case, did you profile your use case already to isolate the bottleneck?

3

The issue is the long extraction time. The model is quite accurate to extract data without errors. But it is taking so much time to do so