bool TensorFlow::doTFLite(int imgHeight, int imgWidth)
{
    Q_UNUSED(imgHeight);
    Q_UNUSED(imgWidth);

    try{
        // Open model & assign error reporter
        model = AuxUtils::getDefaultModelFilename().trimmed().isEmpty() && AuxUtils::getDefaultLabelsFilename().trimmed().isEmpty() ? nullptr :
                FlatBufferModel::BuildFromFile(filename.toStdString().c_str(),&error_reporter);

        if(model == nullptr)
        {
            qDebug() << "TensorFlow model loading: ERROR";
            return false;
        }

        // Link model & resolver
        InterpreterBuilder builder(*model.get(), resolver);

        // Check interpreter
        if(builder(&interpreter) != kTfLiteOk)
        {
            qDebug() << "Interpreter: ERROR";
            return false;
        }

        // Apply accelaration (Neural Network Android)
        
        interpreter->UseNNAPI(accelaration);


        if(interpreter->AllocateTensors() != kTfLiteOk)
        {
            qDebug() << "Allocate tensors: ERROR";
            return false;
        }

        // Set kind of network
        // WARNING: It is assumed the kind of network depending on the number of outputs
        kind_network = interpreter->outputs().size()>1 ? knOBJECT_DETECTION : knIMAGE_SEGMENTATION;

        if (verbose)
        {
          int i_size = interpreter->inputs().size();
          int o_size = interpreter->outputs().size();
          int t_size = interpreter->tensors_size();

          qDebug() << "tensors size: "  << t_size;
          qDebug() << "nodes size: "    << interpreter->nodes_size();
          qDebug() << "inputs: "        << i_size;
          qDebug() << "outputs: "       << o_size;

          for (int i = 0; i < i_size; i++)
            qDebug() << "input" << i << "name:" << interpreter->GetInputName(i) << ", type:" << interpreter->tensor(interpreter->inputs()[i])->type;

          for (int i = 0; i < o_size; i++)
            qDebug() << "output" << i << "name:" << interpreter->GetOutputName(i) << ", type:" << interpreter->tensor(interpreter->outputs()[i])->type;

//          for (int i = 0; i < t_size; i++)
//          {
//            if (interpreter->tensor(i)->name)
//              qDebug()  << i << ":" << interpreter->tensor(i)->name << ","
//                        << interpreter->tensor(i)->bytes << ","
//                        << interpreter->tensor(i)->type << ","
//                        << interpreter->tensor(i)->params.scale << ","
//                        << interpreter->tensor(i)->params.zero_point;
//          }
        }

        // Get input dimension from the input tensor metadata
        // Assuming one input only
        int input = interpreter->inputs()[0];
        TfLiteIntArray* dims = interpreter->tensor(input)->dims;

        // Save outputs
        outputs.clear();
        for(unsigned int i=0;i<interpreter->outputs().size();i++)
            outputs.push_back(interpreter->tensor(interpreter->outputs()[i]));

        wanted_height   = dims->data[1];
        wanted_width    = dims->data[2];
        wanted_channels = dims->data[3];

        if (verbose)
        {
            qDebug() << "Wanted height:"   << wanted_height;
            qDebug() << "Wanted width:"    << wanted_width;
            qDebug() << "Wanted channels:" << wanted_channels;
        }

        if (numThreads > 1)
          interpreter->SetNumThreads(numThreads);

        // Read labels
        if (readLabels()) qDebug() << "There are" << labels.count() << "labels.";
        else qDebug() << "There are NO labels";

        qDebug() << "Tensorflow initialization: OK";
        return true;

    }catch(...)
    {
        qDebug() << "Exception loading model";
        return false;
    }
}

debug output log:
// ====== VPUENC: 4.5.7 build on Mar 11 2021 10:14:03. ======
    // wrapper: 3.0.0 (VPUWRAPPER_ARM64_LINUX Build on Mar 11 2021 09:09:18)
    // vpulib: 1.1.1
    // firmware: 1.1.1.43690
// ERROR: Interpreter::UseNNAPI is not supported. Use Interpreter::ModifyGraphWithDelegate instead.
// tensors size:  185
// nodes size:  70
// inputs:  1
// outputs:  1
// input 0 name: sub_7 , type: 1
// output 0 name: ResizeBilinear_3 , type: 1
// Wanted height: 257
// Wanted width: 257
// Wanted channels: 3
// There are 0 labels.
// Tensorflow initialization: OK
// Elapsed time:  147 milliseconds
// Elapsed time:  182 milliseconds
// Elapsed time:  182 milliseconds
// Elapsed time:  181 milliseconds
// Elapsed time:  181 milliseconds
// Elapsed time:  172 milliseconds