CNNNetwork ReadNetwork(const std::string& model, const) is throwing error as unknown file: Failure

[Subtle]

Hi,

I am using  CNNNetwork ReadNetwork(const std::string& model, const Blob::CPtr& weights) function for loading models from memory.
My code looks like:

//vector<uint8_t> model (.xml)

//vector<uint8_t>weights (.bin)

//for creating TensorDesc

InferenceEngine::TensorDesc tensor(InferenceEngine::Precision::U8,{weights.size()}, InferenceEngine::Layout::ANY);

//For creating Blob

InferenceEngine::TBlob<uint8_t>::Ptr wei_blob = InferenceEngine::make_shared_blob<uint8_t>(tensor,&weights[0],vec_s);

here vec_s is of type size_t

//calling readnetwork 

//std::string strModel(model.begin(),model.end());

InferenceEngine::Core ie;

InferenceEngine::CNNNetwork network =  ie.ReadNetwork(strModel,wei_blob);

My code is getting compiled but when i am running this, it is throwing error :

 unknown file: Failure
C++ exception with description "Failed to construct OpenVINOImageInference" thrown in the test body.

I am not getting what is the issue. 

How can i resolve this.

Thanks a lot

[Zulkifli_Intel]

Hi Subtle,

Thank you for getting in touch.

Which OpenVINO version did you use? We would really appreciate it if you could share the complete file (including the model) for us to replicate and investigate further.

Sincerely,

Zulkifli 

[Subtle]

Hi Zulkifli_intel,

Thanks for replying.

I am using OpenVINO 2021.4 version and OS is Ubuntu 20.04 LTS. I am sorry, i can't share the complete file. I will try to cover needed information in my code snippet here. For my use case i am not going for any particular model. For current scenario i am using face-detection-retail-0004.xml , age-gender-recognition-retail-0013.xml, emotions-recognition-retail-0003.xml with precision FP32 . 

My code looks like(updated):

I have .bin and .xml file stored on disk. I am storing these .bin and .xml to vectors of type uint32_t.

// xmlFile is xmlFile path 

//binFile is binFile path 

string xmlFile;

string binFile;

vector<uint32_t> model (.xml)

vector<uint32_t>weights (.bin)

//storing xmlFile in vector<uint32_t> model

//stroing binFile in vector<uint32_t> weights

//for creating TensorDesc

InferenceEngine::TensorDesc tensor(InferenceEngine::Precision::U32,{weights.size()}, InferenceEngine::Layout::ANY);

//For creating Blob

InferenceEngine::TBlob<uint32_t>::Ptr wei_blob = InferenceEngine::make_shared_blob<uint32_t>(tensor,&weights[0],vec_s);

here vec_s is of type size_t

//calling readnetwork 

//std::string strModel(model.begin(),model.end());

InferenceEngine::Core ie;

InferenceEngine::CNNNetwork network =  ie.ReadNetwork(strModel,wei_blob);

Thank you 

[Zulkifli_Intel]

Hi Subtle,

Sorry for the delay in replying. The "Failed to construct OpenVINOImageInference" error is incomplete. Is there an error message before and after this line? This information may help point to the actual problem you are having.

Sincerely,

Zulkifli 

[Subtle]

Hi Zulkifli_intel,

Thanks for replying.

There is no error message before and after this. 

unknown file: Failure
C++ exception with description "Failed to construct OpenVINOImageInference" thrown in the test body.

I have attached ss for your reference.

Thank you

[Zulkifli_Intel · Accepted solution]

Hi Subtle,

Thank you for sharing the screenshot of the error. Unfortunately, support is hard to come by with your limited information. Is it possible to share your code and steps to reproduce the issue via email? 

Sincerely,

Zulkifli 

[Subtle]

Hi Zulkifli_intel,

Thanks for replying.

Yeah, Sure. I will share artefacts via mail.

Thanks a lot

[Zulkifli_Intel]

Hi Subtle,

You can send it to my email: zulkiflix.bin.abdul.halim@intel.com

Sincerely,

Zulkifli

[Zulkifli_Intel]

Hi Subtle,

By referring to Executable Network, the recommended flow is to read IR content and follow by reading blob content with formatting and allocation before using ReadNetwork to load both.

 // read XML content

std::string xmlString;

std::uint64_t dataSize = 0;

model.read(reinterpret_cast<char\*>(&dataSize), sizeof(dataSize));

xmlString.resize(dataSize);

model.read(const_cast<char\*>(xmlString.c_str()), dataSize);

// read blob content

InferenceEngine::Blob::Ptr dataBlob;

model.read(reinterpret_cast<char\*>(&dataSize), sizeof(dataSize));

if (0 != dataSize) {

dataBlob = InferenceEngine::make_shared_blob<std::uint8_t>(

InferenceEngine::TensorDesc(InferenceEngine::Precision::U8,

{static_cast<std::size_t>(dataSize)},

InferenceEngine::Layout::C));

dataBlob->allocate();

model.read(dataBlob->buffer(), dataSize);

}

auto cnnnetwork = _plugin->GetCore()->ReadNetwork(xmlString, std::move(dataBlob));

In the code, after the blob is created - it was directly fed into ReadNetwork which 'might' potentially caused the error.

// create blob

InferenceEngine::TensorDesc O_tensor(InferenceEngine::Precision::U32,{weights.size()},InferenceEngine::Layout::ANY);

std::cout<<"created tensordesc"<<std::endl;

InferenceEngine::TBlob<uint32_t>::Ptr wei_blob = InferenceEngine::make_shared_blob<uint32_t>(O_tensor,&weights[0]);

std::cout<<"Created blob"<<std::endl;

//Read Network

InferenceEngine::CNNNetwork network = IeCoreSingleton::Instance().ReadNetwork(strModel, wei_blob);

Sincerely,

Zulkifli

[Hari_B_Intel]

Hi Subtle,

Thank you for your patience, after investigating, we found that the model and weight vector are defined as <uint32_t>, and the InferenceEngine::Precision is set to U32. 

So in your code provided, the code was successfully executed after changing the vectors to <unit8_t>, and InferenceEngine::Precision to U8. You can give it a try and see if it works on your program.

std::vector<uint8_t>n_model;

std::vector<uint8_t>weights;

InferenceEngine::TensorDesc O_tensor(InferenceEngine::Precision::U8 {weights.size()},InferenceEngine::Layout::ANY);

InferenceEngine::TBlob<uint8_t>::Ptr wei_blob = InferenceEngine::make_shared_blob<uint8_t>(O_tensor,&weights[0]);

Hope this helps.

Thank you

[Zulkifli_Intel]

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

Sincerely,

Zulkifli