openvino FormatReader 与 cv::Mat 图像 = cv::imdecode

openvino FormatReader vs cv::Mat image = cv::imdecode

我正在尝试将 openvino c++ object_detection_ssd 示例转换为 dll 库,以便我可以在我的 c# 应用程序中使用它。

下面是我导出函数的 C++ 代码

__declspec(dllexport) void Classify_Image(unsigned char* img_pointer, long data_len,
    char* out_result, int length_of_out_result, int top_n_results)
{
    
    std::vector<unsigned char> inputImageBytes(img_pointer, img_pointer + data_len);
    cv::Mat image = cv::imdecode(inputImageBytes,cv::IMREAD_COLOR);
    //cv::imwrite("lala.jpg", image);
    
    if (inputInfo == nullptr) {
        initialize();
    }

    // --------------------------- 8. Create infer request -------------------------------------------------
            //slog::info << "Create infer request" << slog::endl;
    InferRequest infer_request = executable_network.CreateInferRequest();
    // -----------------------------------------------------------------------------------------------------

    std::vector<std::shared_ptr<unsigned char>> imagesData, originalImagesData;
    std::vector<size_t> imageWidths, imageHeights;
    //FormatReader::ReaderPtr reader("C:\Users\Sam\Desktop\la.jpg");
    cv::Mat dst;

    cv::resize(image, dst, cv::Size(400, 225));
    cv::imwrite("lala_m.jpg", dst);
    //std::shared_ptr<unsigned char> originalData(image.data);
    std::shared_ptr<unsigned char> originalData(image.data);

    std::shared_ptr<unsigned char> data1(dst.data);
    //
    originalImagesData.push_back(originalData);
    imagesData.push_back(data1);
    imageWidths.push_back(1280);
    imageHeights.push_back(720);

    if (imagesData.empty()) throw std::logic_error("Valid input images were not found!");

    size_t batchSize = network.getBatchSize();
    
    if (batchSize != imagesData.size()) {
        slog::warn << "Number of images " + std::to_string(imagesData.size()) + \
            " doesn't match batch size " + std::to_string(batchSize) << slog::endl;
        batchSize = std::min(batchSize, imagesData.size());
        slog::warn << "Number of images to be processed is " << std::to_string(batchSize) << slog::endl;
    }
    
    ///** Creating input blob **/
    Blob::Ptr imageInput = infer_request.GetBlob(imageInputName);
    
    ///** Filling input tensor with images. First b channel, then g and r channels **/
    MemoryBlob::Ptr mimage = as<MemoryBlob>(imageInput);
    if (!mimage) {
        slog::err << "We expect image blob to be inherited from MemoryBlob, but by fact we were not able "
            "to cast imageInput to MemoryBlob" << slog::endl;
        return;
    }
    
    //// locked memory holder should be alive all time while access to its buffer happens
    auto minputHolder = mimage->wmap();
    size_t num_channels = mimage->getTensorDesc().getDims()[1];
    size_t image_size = mimage->getTensorDesc().getDims()[3] * mimage->getTensorDesc().getDims()[2];

    unsigned char *data = minputHolder.as<unsigned char *>();

    /** Iterate over all input images **/
    for (size_t image_id = 0; image_id < std::min(imagesData.size(), batchSize); ++image_id) {
        /** Iterate over all pixel in image (b,g,r) **/
        for (size_t pid = 0; pid < image_size; pid++) {
            /** Iterate over all channels **/
            for (size_t ch = 0; ch < num_channels; ++ch) {
                /**          [images stride + channels stride + pixel id ] all in bytes            **/
                data[image_id * image_size * num_channels + ch * image_size + pid] = imagesData.at(image_id).get()[pid*num_channels + ch];
            }
        }
    }

    if (imInfoInputName != "") {
        Blob::Ptr input2 = infer_request.GetBlob(imInfoInputName);
        auto imInfoDim = inputsInfo.find(imInfoInputName)->second->getTensorDesc().getDims()[1];

        /** Fill input tensor with values **/
        MemoryBlob::Ptr minput2 = as<MemoryBlob>(input2);
        if (!minput2) {
            slog::err << "We expect input2 blob to be inherited from MemoryBlob, but by fact we were not able "
                "to cast input2 to MemoryBlob" << slog::endl;
            return;
        }
        // locked memory holder should be alive all time while access to its buffer happens
        auto minput2Holder = minput2->wmap();
        float *p = minput2Holder.as<PrecisionTrait<Precision::FP32>::value_type *>();

        for (size_t image_id = 0; image_id < std::min(imagesData.size(), batchSize); ++image_id) {
            p[image_id * imInfoDim + 0] = static_cast<float>(inputsInfo[imageInputName]->getTensorDesc().getDims()[2]);
            p[image_id * imInfoDim + 1] = static_cast<float>(inputsInfo[imageInputName]->getTensorDesc().getDims()[3]);
            for (size_t k = 2; k < imInfoDim; k++) {
                p[image_id * imInfoDim + k] = 1.0f;  // all scale factors are set to 1.0
            }
        }
    }
    // -----------------------------------------------------------------------------------------------------

    // --------------------------- 10. Do inference ---------------------------------------------------------
    slog::info << "Start inference" << slog::endl;
    infer_request.Infer();
    // -----------------------------------------------------------------------------------------------------

    // --------------------------- 11. Process output -------------------------------------------------------
    slog::info << "Processing output blobs" << slog::endl;

    const Blob::Ptr output_blob = infer_request.GetBlob(outputName);
    MemoryBlob::CPtr moutput = as<MemoryBlob>(output_blob);
    if (!moutput) {
        throw std::logic_error("We expect output to be inherited from MemoryBlob, "
            "but by fact we were not able to cast output to MemoryBlob");
    }
    // locked memory holder should be alive all time while access to its buffer happens
    auto moutputHolder = moutput->rmap();
    const float *detection = moutputHolder.as<const PrecisionTrait<Precision::FP32>::value_type *>();

    std::vector<std::vector<int> > boxes(batchSize);
    std::vector<std::vector<int> > classes(batchSize);

    ///* Each detection has image_id that denotes processed image */
    std::string result = "OB_DATA=";
    int num_detect = 0;
    
    for (int curProposal = 0; curProposal < maxProposalCount; curProposal++) {
        auto image_id = static_cast<int>(detection[curProposal * objectSize + 0]);
        if (image_id < 0 ) {
            slog::info << "ends with break "<<slog::endl;
            break;
        }

        float confidence = detection[curProposal * objectSize + 2];
        auto label = static_cast<int>(detection[curProposal * objectSize + 1]);
        auto xmin = static_cast<int>(detection[curProposal * objectSize + 3] * imageWidths[image_id]);
        auto ymin = static_cast<int>(detection[curProposal * objectSize + 4] * imageHeights[image_id]);
        auto xmax = static_cast<int>(detection[curProposal * objectSize + 5] * imageWidths[image_id]);
        auto ymax = static_cast<int>(detection[curProposal * objectSize + 6] * imageHeights[image_id]);

        std::cout << "[" << curProposal << "," << label << "] element, prob = " << confidence <<
                "    (" << xmin << "," << ymin << ")-(" << xmax << "," << ymax << ")" << " batch id : " << image_id;

        if (confidence > 0.5) {
            num_detect += 1;
            result += std::to_string(confidence);
            
            result += ","+ std::to_string(label);
            result += "," + std::to_string(xmin);
            result += "," + std::to_string(ymin);
            result += "," + std::to_string(xmax);
            result += "," + std::to_string(ymax);
            
            /** Drawing only objects with >50% probability **/
            classes[image_id].push_back(label);
            boxes[image_id].push_back(xmin);
            boxes[image_id].push_back(ymin);
            boxes[image_id].push_back(xmax - xmin);
            boxes[image_id].push_back(ymax - ymin);
            //std::cout << " WILL BE PRINTED!";
            /*std::cout << std::endl;*/
            //slog::info << " add prediction" << slog::endl;
        }
        result += ";";
        std::cout << std::endl;
    }
    
    data1.reset();
    originalData.reset();
    
    //dst.release();
    //image.release();
    

    length_of_out_result = (int)result.size();
    
    std::copy(result.begin(), result.end(), out_result);
    out_result[std::min(length_of_out_result - 1, (int)result.size())] = 0;
            std::cout << "end code"<<std::endl;
    
}

}

在上面的代码中,我试图解码我从 c# 发送到 opencv mat 的字节数组,并 运行 在 mat 对象上进行推理,所以当我从 c# 调用此函数时,它工作正常但打印后“结束代码”行在我的 C# 应用程序中给出了错误,下面是我在 C# 中的代码,我的 C# 应用程序无法打印 Console.WriteLine("after call");在 visual studio 输出中,我可以看到 C++ 函数打印的最后一行是“结束代码”

[DllImport("object_detection_sample_ssd.dll", CallingConvention = CallingConvention.Cdecl)]
    static extern void Classify_Image(byte[] img, long data_len, StringBuilder out_result, int out_result_length, int top_n_results = 2);

    private void button3_Click(object sender, EventArgs e)
    {
        byte[] result = new byte[200];
        Image img = Image.FromFile(@"C:\Users\Sam\Desktop\la.jpg");
        ImageFormat fmt = new ImageFormat(img.RawFormat.Guid);
        var imageCodecInfo = ImageCodecInfo.GetImageEncoders().FirstOrDefault(codec => codec.FormatID == img.RawFormat.Guid);
        //this is for situations, where the image is not read from disk, and is stored in the memort(e.g. image comes from a camera or snapshot)
        if (imageCodecInfo == null)
        {
            fmt = ImageFormat.Jpeg;
        }
        //Image img = Image.FromFile(@"");
        using (MemoryStream ms = new MemoryStream())
        {
            img.Save(ms, fmt);
            byte[] image_byte_array = ms.ToArray();
            int len_result=300;

            int STRING_MAX_LENGTH1 = 300;
            StringBuilder str1 = new StringBuilder(STRING_MAX_LENGTH1);

            Classify_Image(image_byte_array, ms.Length, str1, len_result, 2);
            Console.WriteLine("after call");
        }
        Console.WriteLine("even after using");
        Console.WriteLine("output ="+ ASCIIEncoding.ASCII.GetString(result));
        
    }

我不知道这有什么问题,在 openvino 工具包的原始示例中有一个 FormatReader.h 文件用于从图像文件路径加载图像,我尝试传递图像文件名和使用 FormatReader 作为原始格式,它可以工作。格式阅读器 https://github.com/openvinotoolkit/openvino/tree/master/inference-engine/samples/common/format_reader

请帮忙!

我建议您参考 Hello Classification C++ Sample,它从 Mat 加载图像进行推理。 main.cpp 文件位于以下 link: https://github.com/openvinotoolkit/openvino/blob/master/inference-engine/samples/hello_classification/main.cpp