Method run_RNet

examples/mtcnn/mtcnn.cpp:148–232

}

int mtcnn::run_RNet(image img, std::vector<face_box>& pnet_boxes, std::vector<face_box>& output_boxes)
{
    int batch = pnet_boxes.size();
    int channel = 3;
    int height = 24;
    int width = 24;
    static bool first_run = true;

    tensor_t input_tensor = get_graph_tensor(RNet_graph, "data");
    int dims[] = {batch, channel, height, width};
    set_tensor_shape(input_tensor, dims, 4);
    int img_size = channel * height * width;
    int in_mem = sizeof(float) * batch * img_size;
    float* input_data = ( float* )malloc(in_mem);
    float* input_ptr = input_data;
    set_tensor_buffer(input_tensor, input_ptr, in_mem);

    for(int i = 0; i < batch; i++)
    {
        copy_one_patch(img, pnet_boxes[i], input_ptr, width, height);
        input_ptr += img_size;
    }

    if(first_run)
    {
        if(prerun_graph(RNet_graph) != 0)
        {
            std::cout << "Prerun RNet graph failed, errno: " << get_tengine_errno() << "\n";
            return 1;
        }
        first_run = false;
    }

    if(run_graph(RNet_graph, 1) != 0)
    {
        std::cout << "Run RNet graph failed, errno: " << get_tengine_errno() << "\n";
        return 1;
    }
    free(input_data);
    // std::cout<<"run done ------\n";
    //
    /* output */
    tensor_t tensor = get_graph_tensor(RNet_graph, "conv5-2");
    float* reg_data = ( float* )get_tensor_buffer(tensor);

    tensor = get_graph_tensor(RNet_graph, "prob1");
    float* confidence_data = ( float* )get_tensor_buffer(tensor);

    int conf_page_size = 2;
    int reg_page_size = 4;

    for(int i = 0; i < batch; i++)
    {
        if(*(confidence_data + 1) > conf_r_threshold_)
        {
            face_box output_box;
            face_box& input_box = pnet_boxes[i];