On-premise Setup for Read OCR Container is taking long time to generated OCR for all images.

I. Anand Kumar 0

Hi Team,

For our On-premise OCR engine setup we have used Computer Vision (Read Container) and created OCR container in our local linux machine using the below command:

docker run -d -it -p 5001:5000 --memory 8g --cpus 4 --restart=always mcr.microsoft.com/azure-cognitive-services/vision/read:3.2-model-2022-04-30 Eula=accept Billing=https://gpc-readapi.cognitiveservices.azure.com/ ApiKey= <apikey> HTTP_PROXY= <proxy details> HTTP_PROXY_CREDS= <proxy details>

We are able to generate OCR output for all images. But output performance is taking more time (around 20 - 25 sec per documents) which is creating serious SLA issues from our customers while processing large set of documents (>2k). Can you please review and suggest us to improve the OCR generation performance execution.

Thanks,

Anand Kumar I

YutongTie-MSFT 48,586 Reputation points

2023-08-01T20:18:21.13+00:00

Hello, thanks for reaching out to us. There are several factors that can affect the performance of the Read OCR container, including the size and complexity of the images being processed, the resources allocated to the container, and the network connectivity between the container and the data source.

Here are some suggestions to improve the OCR generation performance:

Increase the resources allocated to the container: You can try increasing the memory and CPU resources allocated to the container to improve its performance. In your docker run command, you have allocated 8GB of memory and 4 CPUs to the container. You can try increasing these values to see if it improves performance.

Optimize the images being processed: The size and complexity of the images being processed can also affect the performance of the OCR engine. You can try optimizing the images by reducing their size or resolution, or by preprocessing them to remove noise or other artifacts that may interfere with OCR.

Use a dedicated network connection: If the container is processing images from a remote data source, network latency and bandwidth can also affect performance. You can try using a dedicated network connection with high bandwidth and low latency to improve performance.

Use a more powerful OCR engine: If the above steps do not improve performance, you may need to consider using a more powerful OCR engine that is designed to handle large volumes of documents. There are several OCR engines available that are optimized for high-volume document processing, such as ABBYY FlexiCapture and Kofax TotalAgility.

I hope it helps, let me know if you have any questions.
I. Anand Kumar 0 Reputation points

2023-08-03T12:03:46.39+00:00

Thanks YutongTie for your prompt response. Will review your suggestions with our internal team and get back to you sooner.
YutongTie-MSFT 48,586 Reputation points

2023-08-06T06:04:57.4533333+00:00

I. Anand Kumar Please let me know how it works, if you need further help on this case, I am willing to enable you a free support ticket to discuss more details.

Regards,

Yutong

Share via

On-premise Setup for Read OCR Container is taking long time to generated OCR for all images.