Yolov5是目标检测中比较经典的模型，学习对其后处理进行解码是非常有必要的。在这里我们仅使用核

函数对Yolov5推理的结果进行解码并恢复成框，掌握后处理所解决的问题，以及对于性能的考虑

经验之谈：

1.对于后处理的代码研究，可以把PyTorch的数据通过转换成numpy后，tobytes再写到文件，然后再到c++中读取的方式，能够快速进行问题研究和排查，此时不需要tensorRT推理也可以做后处理研究。这也叫变量控制法
2.fast_nms_kernel会在极端情况少框，但是这个极端情况一般不会出现，实测几乎没有影响
3.fast nms在cuda实现上比较简单，高效，不用排序

yolov5的输出tensor(n x 85)

其中85是cx, cy, width, height, objness, classification * 80

CPU解码重点：

1.避免多余的计算，需要知道有些数学运算需要的时间远超过很多if，减少他们的次数就是性能的关键
2.nms的实现是可以优化的，例如remove flag并且预先分配内存，reserve对输出分配内存

GPU解码重点：

1.表示输出数量不确定的数组，用[count, box1, box2, box3]的方式，此时需要有最大数量限制
2.通过atomicAdd实现数组元素的加入，并返回索引
3.一样的，不必要的计算，尽量省掉 


#include <cuda_runtime.h>
#include <opencv2/opencv.hpp>
#include <stdio.h>
#include <chrono>
#include <fstream>
#include "box.hpp"

using namespace std;
using namespace cv;

#define checkRuntime(op)  __check_cuda_runtime((op), #op, __FILE__, __LINE__)

bool __check_cuda_runtime(cudaError_t code, const char* op, const char* file, int line){
    if(code != cudaSuccess){
        const char* err_name = cudaGetErrorName(code);    
        const char* err_message = cudaGetErrorString(code);  
        printf("runtime error %s:%d  %s failed. \n  code = %s, message = %s\n", file, line, op, err_name, err_message);   
        return false;
    }
    return true;
}

static std::vector<uint8_t> load_file(const string& file){

    ifstream in(file, ios::in | ios::binary);
    if (!in.is_open())
        return {};

    in.seekg(0, ios::end);
    size_t length = in.tellg();

    std::vector<uint8_t> data;
    if (length > 0){
        in.seekg(0, ios::beg);
        data.resize(length);

        in.read((char*)&data[0], length);
    }
    in.close();
    return data;
}

vector<Box> cpu_decode(float* predict, int rows, int cols, float confidence_threshold = 0.25f, float nms_threshold = 0.45f){
    
    vector<Box> boxes;
    int num_classes = cols - 5;
    for(int i = 0; i < rows; ++i){
        float* pitem = predict + i * cols;
        float objness = pitem[4];
        if(objness < confidence_threshold)
            continue;

        float* pclass = pitem + 5;
        int label     = std::max_element(pclass, pclass + num_classes) - pclass;
        float prob    = pclass[label];
        float confidence = prob * objness;
        if(confidence < confidence_threshold)
            continue;

        float cx     = pitem[0];
        float cy     = pitem[1];
        float width  = pitem[2];
        float height = pitem[3];
        float left   = cx - width * 0.5;
        float top    = cy - height * 0.5;
        float right  = cx + width * 0.5;
        float bottom = cy + height * 0.5;
        boxes.emplace_back(left, top, right, bottom, confidence, (float)label);
    }

    std::sort(boxes.begin(), boxes.end(), [](Box& a, Box& b){return a.confidence > b.confidence;});
    std::vector<bool> remove_flags(boxes.size());
    std::vector<Box> box_result;
    box_result.reserve(boxes.size());

    auto iou = [](const Box& a, const Box& b){
        float cross_left   = std::max(a.left, b.left);
        float cross_top    = std::max(a.top, b.top);
        float cross_right  = std::min(a.right, b.right);
        float cross_bottom = std::min(a.bottom, b.bottom);

        float cross_area = std::max(0.0f, cross_right - cross_left) * std::max(0.0f, cross_bottom - cross_top);
        float union_area = std::max(0.0f, a.right - a.left) * std::max(0.0f, a.bottom - a.top) 
                         + std::max(0.0f, b.right - b.left) * std::max(0.0f, b.bottom - b.top) - cross_area;
        if(cross_area == 0 || union_area == 0) return 0.0f;
        return cross_area / union_area;
    };

    for(int i = 0; i < boxes.size(); ++i){
        if(remove_flags[i]) continue;

        auto& ibox = boxes[i];
        box_result.emplace_back(ibox);
        for(int j = i + 1; j < boxes.size(); ++j){
            if(remove_flags[j]) continue;

            auto& jbox = boxes[j];
            if(ibox.label == jbox.label){
                // class matched
                if(iou(ibox, jbox) >= nms_threshold)
                    remove_flags[j] = true;
            }
        }
    }
    return box_result;
}

void decode_kernel_invoker(
    float* predict, int num_bboxes, int num_classes, float confidence_threshold, 
    float nms_threshold, float* invert_affine_matrix, float* parray, int max_objects, int NUM_BOX_ELEMENT, cudaStream_t stream);

vector<Box> gpu_decode(float* predict, int rows, int cols, float confidence_threshold = 0.25f, float nms_threshold = 0.45f){
    
    vector<Box> box_result;
    cudaStream_t stream = nullptr;
    checkRuntime(cudaStreamCreate(&stream));

    float* predict_device = nullptr;
    float* output_device = nullptr;
    float* output_host = nullptr;
    int max_objects = 1000;
    int NUM_BOX_ELEMENT = 7;  // left, top, right, bottom, confidence, class, keepflag
    checkRuntime(cudaMalloc(&predict_device, rows * cols * sizeof(float)));
    checkRuntime(cudaMalloc(&output_device, sizeof(float) + max_objects * NUM_BOX_ELEMENT * sizeof(float)));
    checkRuntime(cudaMallocHost(&output_host, sizeof(float) + max_objects * NUM_BOX_ELEMENT * sizeof(float)));

    checkRuntime(cudaMemcpyAsync(predict_device, predict, rows * cols * sizeof(float), cudaMemcpyHostToDevice, stream));
    decode_kernel_invoker(
        predict_device, rows, cols - 5, confidence_threshold, 
        nms_threshold, nullptr, output_device, max_objects, NUM_BOX_ELEMENT, stream
    );
    checkRuntime(cudaMemcpyAsync(output_host, output_device, 
        sizeof(int) + max_objects * NUM_BOX_ELEMENT * sizeof(float), 
        cudaMemcpyDeviceToHost, stream
    ));
    checkRuntime(cudaStreamSynchronize(stream));

    int num_boxes = min((int)output_host[0], max_objects);
    for(int i = 0; i < num_boxes; ++i){
        float* ptr = output_host + 1 + NUM_BOX_ELEMENT * i;
        int keep_flag = ptr[6];
        if(keep_flag){
            box_result.emplace_back(
                ptr[0], ptr[1], ptr[2], ptr[3], ptr[4], (int)ptr[5]
            );
        }
    }
    checkRuntime(cudaStreamDestroy(stream));
    checkRuntime(cudaFree(predict_device));
    checkRuntime(cudaFree(output_device));
    checkRuntime(cudaFreeHost(output_host));
    return box_result;
}

int main(){

    auto data = load_file("predict.data");
    auto image = cv::imread("input-image.jpg");
    float* ptr = (float*)data.data();
    int nelem = data.size() / sizeof(float);
    int ncols = 85;
    int nrows = nelem / ncols;
    auto boxes = gpu_decode(ptr, nrows, ncols);
    
    for(auto& box : boxes){
        cv::rectangle(image, cv::Point(box.left, box.top), cv::Point(box.right, box.bottom), cv::Scalar(0, 255, 0), 2);
        cv::putText(image, cv::format("%.2f", box.confidence), cv::Point(box.left, box.top - 7), 0, 0.8, cv::Scalar(0, 0, 255), 2, 16);
    }
    
    cv::imwrite("image-draw.jpg", image);
    return 0;
}

#include <cuda_runtime.h>

static __device__ void affine_project(float* matrix, float x, float y, float* ox, float* oy){
    *ox = matrix[0] * x + matrix[1] * y + matrix[2];
    *oy = matrix[3] * x + matrix[4] * y + matrix[5];
}

static __global__ void decode_kernel(
    float* predict, int num_bboxes, int num_classes, float confidence_threshold, 
    float* invert_affine_matrix, float* parray, int max_objects, int NUM_BOX_ELEMENT
){  
    int position = blockDim.x * blockIdx.x + threadIdx.x;
    if (position >= num_bboxes) return;

    float* pitem     = predict + (5 + num_classes) * position;
    float objectness = pitem[4];
    if(objectness < confidence_threshold)
        return;

    float* class_confidence = pitem + 5;
    float confidence        = *class_confidence++;
    int label               = 0;
    for(int i = 1; i < num_classes; ++i, ++class_confidence){
        if(*class_confidence > confidence){
            confidence = *class_confidence;
            label      = i;
        }
    }

    confidence *= objectness;
    if(confidence < confidence_threshold)
        return;

    int index = atomicAdd(parray, 1);
    if(index >= max_objects)
        return;

    float cx         = *pitem++;
    float cy         = *pitem++;
    float width      = *pitem++;
    float height     = *pitem++;
    float left   = cx - width * 0.5f;
    float top    = cy - height * 0.5f;
    float right  = cx + width * 0.5f;
    float bottom = cy + height * 0.5f;
    // affine_project(invert_affine_matrix, left,  top,    &left,  &top);
    // affine_project(invert_affine_matrix, right, bottom, &right, &bottom);

    // left, top, right, bottom, confidence, class, keepflag
    float* pout_item = parray + 1 + index * NUM_BOX_ELEMENT;
    *pout_item++ = left;
    *pout_item++ = top;
    *pout_item++ = right;
    *pout_item++ = bottom;
    *pout_item++ = confidence;
    *pout_item++ = label;
    *pout_item++ = 1; // 1 = keep, 0 = ignore
}

static __device__ float box_iou(
    float aleft, float atop, float aright, float abottom, 
    float bleft, float btop, float bright, float bbottom
){

    float cleft 	= max(aleft, bleft);
    float ctop 		= max(atop, btop);
    float cright 	= min(aright, bright);
    float cbottom 	= min(abottom, bbottom);
    
    float c_area = max(cright - cleft, 0.0f) * max(cbottom - ctop, 0.0f);
    if(c_area == 0.0f)
        return 0.0f;
    
    float a_area = max(0.0f, aright - aleft) * max(0.0f, abottom - atop);
    float b_area = max(0.0f, bright - bleft) * max(0.0f, bbottom - btop);
    return c_area / (a_area + b_area - c_area);
}

static __global__ void fast_nms_kernel(float* bboxes, int max_objects, float threshold, int NUM_BOX_ELEMENT){

    int position = (blockDim.x * blockIdx.x + threadIdx.x);
    int count = min((int)*bboxes, max_objects);
    if (position >= count) 
        return;
    
    // left, top, right, bottom, confidence, class, keepflag
    float* pcurrent = bboxes + 1 + position * NUM_BOX_ELEMENT;
    for(int i = 0; i < count; ++i){
        float* pitem = bboxes + 1 + i * NUM_BOX_ELEMENT;
        if(i == position || pcurrent[5] != pitem[5]) continue;

        if(pitem[4] >= pcurrent[4]){
            if(pitem[4] == pcurrent[4] && i < position)
                continue;

            float iou = box_iou(
                pcurrent[0], pcurrent[1], pcurrent[2], pcurrent[3],
                pitem[0],    pitem[1],    pitem[2],    pitem[3]
            );

            if(iou > threshold){
                pcurrent[6] = 0;  // 1=keep, 0=ignore
                return;
            }
        }
    }
} 

void decode_kernel_invoker(
    float* predict, int num_bboxes, int num_classes, float confidence_threshold, 
    float nms_threshold, float* invert_affine_matrix, float* parray, int max_objects, int NUM_BOX_ELEMENT, cudaStream_t stream){
    
    auto block = num_bboxes > 512 ? 512 : num_bboxes;
    auto grid = (num_bboxes + block - 1) / block;

    /* 如果核函数有波浪线，没关系，他是正常的，你只是看不顺眼罢了 */
    decode_kernel<<<grid, block, 0, stream>>>(
        predict, num_bboxes, num_classes, confidence_threshold, 
        invert_affine_matrix, parray, max_objects, NUM_BOX_ELEMENT
    );

    block = max_objects > 512 ? 512 : max_objects;
    grid = (max_objects + block - 1) / block;
    fast_nms_kernel<<<grid, block, 0, stream>>>(parray, max_objects, nms_threshold, NUM_BOX_ELEMENT);
}

# pytorch # 深度学习 # 神经网络
Logo
天工造物局

更多推荐

肌肉骨骼仿真如何重塑机器人控制？探索MyoSuite的强化学习革命

想象一下，让机器人像人类一样自然地抓取物体、行走、甚至打乒乓球——这不再是科幻电影中的场景。MyoSuite，这个基于MuJoCo物理引擎和OpenAI Gym API的开源肌肉骨骼仿真套件，正在重新定义机器人控制的可能性。在过去的几年里，强化学习在机器人控制领域取得了显著进展，但大多数研究都集中在刚性机器人上。MyoSuite则专注于更接近生物系统的肌肉骨骼模型，为研究者提供了一个探索生物力学控

avatar 天工造物局
cover

具身智能数据采集：从实验室到生活，核数聚如何为机器人 “喂饱” 真实数据

avatar 天工造物局

[具身智能-789]：Gazebo + RViz2 + 控制算法+ROS2：，构建一套完全数字化的机器人学习和监控系统

创造虚拟世界，产生运动与传感器数据（数据源）控制算法：做决策，输出运动指令（大脑）RViz2：可视化观测，学习与监控窗口（观察者）

avatar 天工造物局