ncnn_pnnx初探

1、训练代码

# 训练我家小喵咪nora的照片
# 用的是yolov5 v6.1版本的ghost网络-->/models/hub/yolov5s-ghost.yaml
# 训练请自己百度哈哈

2、导出torchscript

#在yolov5修改export.py，导出torchscript模型，修改要添加模型为训练模式 即model.train()

3、下载pnnx或者自己编译pnnx

这里我在window下测试了，一个是直接下载编译好的pnnx.exe,另外一个是下载官网源码，参考：https://zhuanlan.zhihu.com/p/431833958编译了window版本，出现了转换不报错，但是什么都没生成的结果(param bin等都没生成)，于是转换到linux下转换，我的是centos7，所以我在centos下重新编译和生成。

注意注意：本人电脑是cpu版本，所以要在export.py中添加cpu版本转换。具体加如下：

成功生成

4、查看nora.ncnn.param，找出输出的3个output，分别为out0，out1,out2

5、用vs测试一下，依旧是拿nihui的demo测试

#include "layer.h"
#include "net.h"

#if defined(USE_NCNN_SIMPLEOCV)
#include "simpleocv.h"
#else
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc/imgproc.hpp>
#endif
#include <float.h>
#include <stdio.h>
#include <vector>

struct Object
{
    cv::Rect_<float> rect;
    int label;
    float prob;
};

static inline float intersection_area(const Object& a, const Object& b)
{
    cv::Rect_<float> inter = a.rect & b.rect;
    return inter.area();
}

static void qsort_descent_inplace(std::vector<Object>& faceobjects, int left, int right)
{
    int i = left;
    int j = right;
    float p = faceobjects[(left + right) / 2].prob;

    while (i <= j)
    {
        while (faceobjects[i].prob > p)
            i++;

        while (faceobjects[j].prob < p)
            j--;

        if (i <= j)
        {
            // swap
            std::swap(faceobjects[i], faceobjects[j]);

            i++;
            j--;
        }
    }

#pragma omp parallel sections
    {
#pragma omp section
        {
            if (left < j) qsort_descent_inplace(faceobjects, left, j);
        }
#pragma omp section
        {
            if (i < right) qsort_descent_inplace(faceobjects, i, right);
        }
    }
}

static void qsort_descent_inplace(std::vector<Object>& faceobjects)
{
    if (faceobjects.empty())
        return;

    qsort_descent_inplace(faceobjects, 0, faceobjects.size() - 1);
}

static void nms_sorted_bboxes(const std::vector<Object>& faceobjects, std::vector<int>& picked, float nms_threshold)
{
    picked.clear();

    const int n = faceobjects.size();

    std::vector<float> areas(n);
    for (int i = 0; i < n; i++)
    {
        areas[i] = faceobjects[i].rect.area();
    }

    for (int i = 0; i < n; i++)
    {
        const Object& a = faceobjects[i];

        int keep = 1;
        for (int j = 0; j < (int)picked.size(); j++)
        {
            const Object& b = faceobjects[picked[j]];

            // intersection over union
            float inter_area = intersection_area(a, b);
            float union_area = areas[i] + areas[picked[j]] - inter_area;
            // float IoU = inter_area / union_area
            if (inter_area / union_area > nms_threshold)
                keep = 0;
        }

        if (keep)
            picked.push_back(i);
    }
}

static inline float sigmoid(float x)
{
    return static_cast<float>(1.f / (1.f + exp(-x)));
}

static void generate_proposals(const ncnn::Mat& anchors, int stride, const ncnn::Mat& in_pad, const ncnn::Mat& feat_blob, float prob_threshold, std::vector<Object>& objects)
{
    const int num_grid_x = feat_blob.w;
    const int num_grid_y = feat_blob.h;

    const int num_anchors = anchors.w / 2;

    const int num_class = feat_blob.c / num_anchors - 5;

    const int feat_offset = num_class + 5;

    for (int q = 0; q < num_anchors; q++)
    {
        const float anchor_w = anchors[q * 2];
        const float anchor_h = anchors[q * 2 + 1];

        for (int i = 0; i < num_grid_y; i++)
        {
            for (int j = 0; j < num_grid_x; j++)
            {
                // find class index with max class score
                int class_index = 0;
                float class_score = -FLT_MAX;
                for (int k = 0; k < num_class; k++)
                {
                    float score = feat_blob.channel(q * feat_offset + 5 + k).row(i)[j];
                    if (score > class_score)
                    {
                        class_index = k;
                        class_score = score;
                    }
                }

                float box_score = feat_blob.channel(q * feat_offset + 4).row(i)[j];

                float confidence = sigmoid(box_score) * sigmoid(class_score);

                if (confidence >= prob_threshold)
                {
                    // yolov5/models/yolo.py Detect forward
                    // y = x[i].sigmoid()
                    // y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i].to(x[i].device)) * self.stride[i]  # xy
                    // y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh

                    float dx = sigmoid(feat_blob.channel(q * feat_offset + 0).row(i)[j]);
                    float dy = sigmoid(feat_blob.channel(q * feat_offset + 1).row(i)[j]);
                    float dw = sigmoid(feat_blob.channel(q * feat_offset + 2).row(i)[j]);
                    float dh = sigmoid(feat_blob.channel(q * feat_offset + 3).row(i)[j]);

                    float pb_cx = (dx * 2.f - 0.5f + j) * stride;
                    float pb_cy = (dy * 2.f - 0.5f + i) * stride;

                    float pb_w = pow(dw * 2.f, 2) * anchor_w;
                    float pb_h = pow(dh * 2.f, 2) * anchor_h;

                    float x0 = pb_cx - pb_w * 0.5f;
                    float y0 = pb_cy - pb_h * 0.5f;
                    float x1 = pb_cx + pb_w * 0.5f;
                    float y1 = pb_cy + pb_h * 0.5f;

                    Object obj;
                    obj.rect.x = x0;
                    obj.rect.y = y0;
                    obj.rect.width = x1 - x0;
                    obj.rect.height = y1 - y0;
                    obj.label = class_index;
                    obj.prob = confidence;

                    objects.push_back(obj);
                }
            }
        }
    }
}

static int detect_yolov5(const cv::Mat& bgr, std::vector<Object>& objects)
{
    ncnn::Net yolov5;

    yolov5.opt.use_vulkan_compute = true;
    // yolov5.opt.use_bf16_storage = true;

    // original pretrained model from https://github.com/ultralytics/yolov5
    // the ncnn model https://github.com/nihui/ncnn-assets/tree/master/models
    yolov5.load_param("E:\\c++Project\\202205\\yolov5_android_vs2019\\ghostyolo\\nora.ncnn.param");
    yolov5.load_model("E:\\c++Project\\202205\\yolov5_android_vs2019\\ghostyolo\\nora.ncnn.bin");

    const int target_size = 416;
    const float prob_threshold = 0.25f;
    const float nms_threshold = 0.45f;

    int img_w = bgr.cols;
    int img_h = bgr.rows;

    // yolov5/models/common.py DetectMultiBackend
    const int max_stride = 64;

    // letterbox pad to multiple of max_stride
    int w = img_w;
    int h = img_h;
    float scale = 1.f;
    if (w > h)
    {
        scale = (float)target_size / w;
        w = target_size;
        h = h * scale;
    }
    else
    {
        scale = (float)target_size / h;
        h = target_size;
        w = w * scale;
    }

    ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR2RGB, img_w, img_h, w, h);

    // pad to target_size rectangle
    // yolov5/utils/datasets.py letterbox
    int wpad = (w + max_stride - 1) / max_stride * max_stride - w;
    int hpad = (h + max_stride - 1) / max_stride * max_stride - h;
    ncnn::Mat in_pad;
    ncnn::copy_make_border(in, in_pad, hpad / 2, hpad - hpad / 2, wpad / 2, wpad - wpad / 2, ncnn::BORDER_CONSTANT, 114.f);

    const float norm_vals[3] = { 1 / 255.f, 1 / 255.f, 1 / 255.f };
    in_pad.substract_mean_normalize(0, norm_vals);

    ncnn::Extractor ex = yolov5.create_extractor();

    ex.input("in0", in_pad);

    std::vector<Object> proposals;

    // anchor setting from yolov5/models/yolov5s.yaml

    // stride 8
    {
        ncnn::Mat out;
        ex.extract("out0", out);

        ncnn::Mat anchors(6);
        anchors[0] = 10.f;
        anchors[1] = 13.f;
        anchors[2] = 16.f;
        anchors[3] = 30.f;
        anchors[4] = 33.f;
        anchors[5] = 23.f;

        std::vector<Object> objects8;
        generate_proposals(anchors, 8, in_pad, out, prob_threshold, objects8);

        proposals.insert(proposals.end(), objects8.begin(), objects8.end());
    }

    // stride 16
    {
        ncnn::Mat out;
        ex.extract("out1", out);

        ncnn::Mat anchors(6);
        anchors[0] = 30.f;
        anchors[1] = 61.f;
        anchors[2] = 62.f;
        anchors[3] = 45.f;
        anchors[4] = 59.f;
        anchors[5] = 119.f;

        std::vector<Object> objects16;
        generate_proposals(anchors, 16, in_pad, out, prob_threshold, objects16);

        proposals.insert(proposals.end(), objects16.begin(), objects16.end());
    }

    // stride 32
    {
        ncnn::Mat out;
        ex.extract("out2", out);

        ncnn::Mat anchors(6);
        anchors[0] = 116.f;
        anchors[1] = 90.f;
        anchors[2] = 156.f;
        anchors[3] = 198.f;
        anchors[4] = 373.f;
        anchors[5] = 326.f;

        std::vector<Object> objects32;
        generate_proposals(anchors, 32, in_pad, out, prob_threshold, objects32);

        proposals.insert(proposals.end(), objects32.begin(), objects32.end());
    }

    // sort all proposals by score from highest to lowest
    qsort_descent_inplace(proposals);

    // apply nms with nms_threshold
    std::vector<int> picked;
    nms_sorted_bboxes(proposals, picked, nms_threshold);

    int count = picked.size();

    objects.resize(count);
    for (int i = 0; i < count; i++)
    {
        objects[i] = proposals[picked[i]];

        // adjust offset to original unpadded
        float x0 = (objects[i].rect.x - (wpad / 2)) / scale;
        float y0 = (objects[i].rect.y - (hpad / 2)) / scale;
        float x1 = (objects[i].rect.x + objects[i].rect.width - (wpad / 2)) / scale;
        float y1 = (objects[i].rect.y + objects[i].rect.height - (hpad / 2)) / scale;

        // clip
        x0 = std::max(std::min(x0, (float)(img_w - 1)), 0.f);
        y0 = std::max(std::min(y0, (float)(img_h - 1)), 0.f);
        x1 = std::max(std::min(x1, (float)(img_w - 1)), 0.f);
        y1 = std::max(std::min(y1, (float)(img_h - 1)), 0.f);

        objects[i].rect.x = x0;
        objects[i].rect.y = y0;
        objects[i].rect.width = x1 - x0;
        objects[i].rect.height = y1 - y0;
    }

    return 0;
}

static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects)
{
    static const char* class_names[] = {
        "nora"
    };

    cv::Mat image = bgr.clone();

    for (size_t i = 0; i < objects.size(); i++)
    {
        const Object& obj = objects[i];

        fprintf(stderr, "%d = %.5f at %.2f %.2f %.2f x %.2f\n", obj.label, obj.prob,
            obj.rect.x, obj.rect.y, obj.rect.width, obj.rect.height);

        cv::rectangle(image, obj.rect, cv::Scalar(255, 0, 0));

        char text[256];
        sprintf_s(text, "%s %.1f%%", class_names[obj.label], obj.prob * 100);

        int baseLine = 0;
        cv::Size label_size = cv::getTextSize(text, cv::FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);

        int x = obj.rect.x;
        int y = obj.rect.y - label_size.height - baseLine;
        if (y < 0)
            y = 0;
        if (x + label_size.width > image.cols)
            x = image.cols - label_size.width;

        cv::rectangle(image, cv::Rect(cv::Point(x, y), cv::Size(label_size.width, label_size.height + baseLine)),
            cv::Scalar(255, 255, 255), -1);

        cv::putText(image, text, cv::Point(x, y + label_size.height),
            cv::FONT_HERSHEY_SIMPLEX, 0.5, cv::Scalar(0, 0, 0));
    }

    cv::imwrite("E:/c++Project/202205/yolov5_android_vs2019/yolov5.jpg", image);
}

int main(int argc, char** argv)
{
   /* if (argc != 2)
    {
        fprintf(stderr, "Usage: %s [imagepath]\n", argv[0]);
        return -1;
    }*/
    std::string imagepath = "E:/c++Project/202205/yolov5_android_vs2019/1.jpg";

    cv::Mat m = cv::imread(imagepath.c_str(), 1);
    if (m.empty())
    {
        fprintf(stderr, "cv::imread %s failed\n", imagepath);
        return -1;
    }

    std::vector<Object> objects;
    detect_yolov5(m, objects);

    draw_objects(m, objects);

    return 0;
}

6、结果