Using Inference Engine ReadNetwork method to load models from memory

[Subtle]

Hi,

I want OpenVINO’s inference engine to load models from memory (not from a file path).

For this I am using  CNNNetwork ReadNetwork(const std::string& model, const Blob::CPtr& weights) function.

I need help regarding how can i create Blob::CPtr weights from weights.bin file. I am using OpenVINO 2021.4 version and OS is Ubuntu 20.04 LTS.

Thank you 

[Peh_Intel · Accepted solution]

Hi Subtle,

You can create Blob objects using constructors with InferenceEngine::TensorDesc. You can refer to Inference Memory Primitives (C++) for more information.

Besides, I also found that someone shared his way to load model from memory in the previous post which may be helpful for you.

Regards,

Peh

[Subtle]

Hi Peh,

Thanks for reply. 

I am creating Blob objects using constructor  TensorDesc(const Precision& precision, const SizeVector& dims, Layout layout) . For my use case i need to keep Precision::UNSPECIFIED and Layout::ANY, I want to understand more about const SizeVector& dims, what should i keep for this parameter, if i am keeping Layout::ANY. 

Thank you 

[Peh_Intel]

Hi Subtle,

There is nothing special for the layout set as “ANY”. The parameters for the dimension still need to set based on the model in the supported format (InferenceEngine::Layout::NCDHW, InferenceEngine::Layout::NCHW, InferenceEngine::Layout::NC, InferenceEngine::Layout::C).

Regards,

Peh

[Subtle]

Hi Peh, 

Thanks a lot for reply. 

I got the point. 

Thanks & Regards 

Subtle 

[Peh_Intel]

Hi Subtle,

This thread will no longer be monitored since we have provided answers. If you need any additional information from Intel, please submit a new question. 

Regards,

Peh