使用OpenCV 4.1.2的DNN模块部署深度学习模型

int main()
{
    float model = "ssdlite_mobilenet_v2.pb";
    float config = "ssdlite_mobilenet_v2.pbtxt";
    float scale = 1.0;
    int inpWidth = 300;
    int inpHeight = 300;
    Scalar mean(103.939, 116.779, 123.68);
    float confThreshold = 0.5f;
    float nmsThreshold = 0.4f;
    // 1. 装载模型，设置参数
    Net net = readNet(model, config);
    net.setPreferableBackend(DNN_BACKEND_OPENCV);
    net.setPreferableTarget(DNN_TARGET_CPU);
    std::vector<String> outNames = net.getUnconnectedOutLayersNames();

    // 读入图像
    Mat image = imread("picture.jpg");

    Mat blob;
    // 2. 由输入图像生成blob，设置blob为模型输入
    blobFromImage(image, blob, scale, inpSize, mean, true, false);
    net.setInput(blob);
    std::vector<Mat> outs;
    // 3. 前向计算，检测结果存入outs
    net.forward(outs, outNames);
    // 4. 对检测结果进行非极大值抑制等后处理，结果存入outs
    postprocess(frame, outs, net);
}

int main()
{
    float model = "ssdlite_mobilenet_v2.pb";
    float config = "ssdlite_mobilenet_v2.pbtxt";
    float scale = 1.0;
    int inpWidth = 300;
    int inpHeight = 300;
    Scalar mean(103.939, 116.779, 123.68);
    float confThreshold = 0.5f;
    float nmsThreshold = 0.4f;
    // 1. 装载模型，设置参数
    DetectionModel dnn_model(model, config);
    dnn_model.setPreferableBackend(DNN_BACKEND_OPENCV);
    dnn_model.setPreferableTarget(DNN_TARGET_CPU);
    dnn_model.setInputParams(scale, Size(inpWidth, inpHeight), mean, true, false);

    // 读入图像
    Mat image = imread("picture.jpg");

    std::vector<int> classIds;
    std::vector<float> confs;
    std::vector<Rect> boxes;
    // 2. 前向计算，检测结果矩形框存入boxes（已经过后处理）
    dnn_model.detect(image, classIds, confs, boxes, confThreshold, nmsThreshold);
}