Deploy parts of OpenVINO required for executing Inference on pre-trained Models as AWS Lambda Layer

[SRado1]

Hello!

I have code written in Go Programming Language which uses models trained with OpenVINO framework (.xml and .bin files).

I would like to know, which parts of OpenVINO do I need to package into AWS Lambda Layer so that I can perform Inference with already trained models?

Thing is, OpenVINO takes ~900MB and AWS Lambda Function maximum size is 250MB. I need to take ONLY PARTS of OpenVINO that allow me to perform Inference over pre-trained DNN models (xml and bin files) and package them as AWS Lambda Layer.

Thanks in advance!

[Artem_A_Intel]

Hello!

Starting with OpenVINO 2019'R3 release you can use Deployment Manager tool for this purpose.

This tool allows to generate deployment packages with necessary target for inference(CPU/GPU/etc) and user data - in your case it is Go application and models trained with OpenVINO framework (.xml and .bin files).

Please find more details in documentation
https://docs.openvinotoolkit.org/2019_R3/_docs_install_guides_deployment_manager_tool.html

[Sundar__Sharan]

Deployment Packages do not support python bindings right? I have tried and it always reports "openvino module not found". But AWS Lambda supports only Python and several other runtimes except C++. So deployment package does not work here for me. Any other alternative for deploying using python with dependencies size less than 250mb?