diff --git a/yolov5/CMakeLists.txt b/yolov5/CMakeLists.txt
new file mode 100644
index 0000000..55f84c9
--- /dev/null
+++ b/yolov5/CMakeLists.txt
@@ -0,0 +1,41 @@
+cmake_minimum_required(VERSION 2.6)
+
+project(yolov5)
+
+add_definitions(-std=c++11)
+
+option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_BUILD_TYPE Debug)
+
+find_package(CUDA REQUIRED)
+
+set(CUDA_NVCC_PLAGS ${CUDA_NVCC_PLAGS};-std=c++11;-g;-G;-gencode;arch=compute_30;code=sm_30)
+
+include_directories(${PROJECT_SOURCE_DIR}/include)
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+    message("embed_platform on")
+    include_directories(/usr/local/cuda/targets/aarch64-linux/include)
+    link_directories(/usr/local/cuda/targets/aarch64-linux/lib)
+else()
+    message("embed_platform off")
+    include_directories(/usr/local/cuda/include)
+    link_directories(/usr/local/cuda/lib64)
+endif()
+
+
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Ofast -Wfatal-errors -D_MWAITXINTRIN_H_INCLUDED")
+
+cuda_add_library(yololayer SHARED ${PROJECT_SOURCE_DIR}/yololayer.cu)
+
+find_package(OpenCV)
+include_directories(OpenCV_INCLUDE_DIRS)
+
+add_executable(yolov5s ${PROJECT_SOURCE_DIR}/yolov5s.cpp)
+target_link_libraries(yolov5s nvinfer)
+target_link_libraries(yolov5s cudart)
+target_link_libraries(yolov5s yololayer)
+target_link_libraries(yolov5s ${OpenCV_LIBS})
+
+add_definitions(-O2 -pthread)
+
diff --git a/yolov5/README.md b/yolov5/README.md
new file mode 100644
index 0000000..eb900b7
--- /dev/null
+++ b/yolov5/README.md
@@ -0,0 +1,52 @@
+# yolov5
+
+The Pytorch implementation is [ultralytics/yolov5](https://github.com/ultralytics/yolov5).
+
+## How to Run
+
+```
+1. generate yolov5s.wts from pytorch implementation with yolov5s.pt
+
+git clone https://github.com/wang-xinyu/tensorrtx.git
+git clone https://github.com/ultralytics/yolov5.git
+// download its weights 'yolov5s.pt'
+cd yolov5
+cp ../tensorrtx/yolov5s/gen_wts.py .
+python gen_wts.py
+// a file 'yolov5s.wts' will be generated.
+
+2. put yolov5s.wts into yolov5, build and run
+
+mv yolov5s.wts ../tensorrtx/yolov5/
+cd ../tensorrtx/yolov5
+mkdir build
+cd build
+cmake ..
+make
+sudo ./yolov5s -s             // serialize model to plan file i.e. 'yolov5s.engine'
+sudo ./yolov5s -d  ../samples // deserialize plan file and run inference, the images in samples will be processed.
+
+3. check the images generated, as follows. _zidane.jpg and _bus.jpg
+```
+
+<p align="center">
+<img src="https://user-images.githubusercontent.com/15235574/78247927-4d9fac00-751e-11ea-8b1b-704a0aeb3fcf.jpg">
+</p>
+
+<p align="center">
+<img src="https://user-images.githubusercontent.com/15235574/78247970-60b27c00-751e-11ea-88df-41473fed4823.jpg">
+</p>
+
+## Config
+
+- Input shape defined in yololayer.h
+- Number of classes defined in yololayer.h
+- FP16/FP32 can be selected by the macro in yolov5s.cpp
+- GPU id can be selected by the macro in yolov5s.cpp
+- NMS thresh in yolov5s.cpp
+- BBox confidence thresh in yolov5s.cpp
+
+## More Information
+
+See the readme in [home page.](https://github.com/wang-xinyu/tensorrtx)
+
diff --git a/yolov5/common.hpp b/yolov5/common.hpp
new file mode 100644
index 0000000..20ce35c
--- /dev/null
+++ b/yolov5/common.hpp
@@ -0,0 +1,295 @@
+#ifndef YOLOV5_COMMON_H_
+#define YOLOV5_COMMON_H_
+
+#include <fstream>
+#include <map>
+#include <sstream>
+#include <vector>
+#include <opencv2/opencv.hpp>
+#include <dirent.h>
+#include "NvInfer.h"
+#include "yololayer.h"
+
+#define CHECK(status) \
+    do\
+    {\
+        auto ret = (status);\
+        if (ret != 0)\
+        {\
+            std::cerr << "Cuda failure: " << ret << std::endl;\
+            abort();\
+        }\
+    } while (0)
+
+using namespace nvinfer1;
+
+cv::Mat preprocess_img(cv::Mat& img) {
+    int w, h, x, y;
+    float r_w = Yolo::INPUT_W / (img.cols*1.0);
+    float r_h = Yolo::INPUT_H / (img.rows*1.0);
+    if (r_h > r_w) {
+        w = Yolo::INPUT_W;
+        h = r_w * img.rows;
+        x = 0;
+        y = (Yolo::INPUT_H - h) / 2;
+    } else {
+        w = r_h* img.cols;
+        h = Yolo::INPUT_H;
+        x = (Yolo::INPUT_W - w) / 2;
+        y = 0;
+    }
+    cv::Mat re(h, w, CV_8UC3);
+    cv::resize(img, re, re.size(), 0, 0, cv::INTER_CUBIC);
+    cv::Mat out(Yolo::INPUT_H, Yolo::INPUT_W, CV_8UC3, cv::Scalar(128, 128, 128));
+    re.copyTo(out(cv::Rect(x, y, re.cols, re.rows)));
+    return out;
+}
+
+cv::Rect get_rect(cv::Mat& img, float bbox[4]) {
+    int l, r, t, b;
+    float r_w = Yolo::INPUT_W / (img.cols * 1.0);
+    float r_h = Yolo::INPUT_H / (img.rows * 1.0);
+    if (r_h > r_w) {
+        l = bbox[0] - bbox[2]/2.f;
+        r = bbox[0] + bbox[2]/2.f;
+        t = bbox[1] - bbox[3]/2.f - (Yolo::INPUT_H - r_w * img.rows) / 2;
+        b = bbox[1] + bbox[3]/2.f - (Yolo::INPUT_H - r_w * img.rows) / 2;
+        l = l / r_w;
+        r = r / r_w;
+        t = t / r_w;
+        b = b / r_w;
+    } else {
+        l = bbox[0] - bbox[2]/2.f - (Yolo::INPUT_W - r_h * img.cols) / 2;
+        r = bbox[0] + bbox[2]/2.f - (Yolo::INPUT_W - r_h * img.cols) / 2;
+        t = bbox[1] - bbox[3]/2.f;
+        b = bbox[1] + bbox[3]/2.f;
+        l = l / r_h;
+        r = r / r_h;
+        t = t / r_h;
+        b = b / r_h;
+    }
+    return cv::Rect(l, t, r-l, b-t);
+}
+
+float iou(float lbox[4], float rbox[4]) {
+    float interBox[] = {
+        std::max(lbox[0] - lbox[2]/2.f , rbox[0] - rbox[2]/2.f), //left
+        std::min(lbox[0] + lbox[2]/2.f , rbox[0] + rbox[2]/2.f), //right
+        std::max(lbox[1] - lbox[3]/2.f , rbox[1] - rbox[3]/2.f), //top
+        std::min(lbox[1] + lbox[3]/2.f , rbox[1] + rbox[3]/2.f), //bottom
+    };
+
+    if(interBox[2] > interBox[3] || interBox[0] > interBox[1])
+        return 0.0f;
+
+    float interBoxS =(interBox[1]-interBox[0])*(interBox[3]-interBox[2]);
+    return interBoxS/(lbox[2]*lbox[3] + rbox[2]*rbox[3] -interBoxS);
+}
+
+bool cmp(Yolo::Detection& a, Yolo::Detection& b) {
+    return a.conf > b.conf;
+}
+
+void nms(std::vector<Yolo::Detection>& res, float *output, float conf_thresh, float nms_thresh = 0.5) {
+    int det_size = sizeof(Yolo::Detection) / sizeof(float);
+    std::map<float, std::vector<Yolo::Detection>> m;
+    for (int i = 0; i < output[0] && i < 1000; i++) {
+        if (output[1 + det_size * i + 4] <= conf_thresh) continue;
+        Yolo::Detection det;
+        memcpy(&det, &output[1 + det_size * i], det_size * sizeof(float));
+        if (m.count(det.class_id) == 0) m.emplace(det.class_id, std::vector<Yolo::Detection>());
+        m[det.class_id].push_back(det);
+    }
+    for (auto it = m.begin(); it != m.end(); it++) {
+        //std::cout << it->second[0].class_id << " --- " << std::endl;
+        auto& dets = it->second;
+        std::sort(dets.begin(), dets.end(), cmp);
+        for (size_t m = 0; m < dets.size(); ++m) {
+            auto& item = dets[m];
+            res.push_back(item);
+            for (size_t n = m + 1; n < dets.size(); ++n) {
+                if (iou(item.bbox, dets[n].bbox) > nms_thresh) {
+                    dets.erase(dets.begin()+n);
+                    --n;
+                }
+            }
+        }
+    }
+}
+
+// TensorRT weight files have a simple space delimited format:
+// [type] [size] <data x size in hex>
+std::map<std::string, Weights> loadWeights(const std::string file) {
+    std::cout << "Loading weights: " << file << std::endl;
+    std::map<std::string, Weights> weightMap;
+
+    // Open weights file
+    std::ifstream input(file);
+    assert(input.is_open() && "Unable to load weight file.");
+
+    // Read number of weight blobs
+    int32_t count;
+    input >> count;
+    assert(count > 0 && "Invalid weight map file.");
+
+    while (count--)
+    {
+        Weights wt{DataType::kFLOAT, nullptr, 0};
+        uint32_t size;
+
+        // Read name and type of blob
+        std::string name;
+        input >> name >> std::dec >> size;
+        wt.type = DataType::kFLOAT;
+
+        // Load blob
+        uint32_t* val = reinterpret_cast<uint32_t*>(malloc(sizeof(val) * size));
+        for (uint32_t x = 0, y = size; x < y; ++x)
+        {
+            input >> std::hex >> val[x];
+        }
+        wt.values = val;
+        
+        wt.count = size;
+        weightMap[name] = wt;
+    }
+
+    return weightMap;
+}
+
+IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, std::string lname, float eps) {
+    float *gamma = (float*)weightMap[lname + ".weight"].values;
+    float *beta = (float*)weightMap[lname + ".bias"].values;
+    float *mean = (float*)weightMap[lname + ".running_mean"].values;
+    float *var = (float*)weightMap[lname + ".running_var"].values;
+    int len = weightMap[lname + ".running_var"].count;
+
+    float *scval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
+    for (int i = 0; i < len; i++) {
+        scval[i] = gamma[i] / sqrt(var[i] + eps);
+    }
+    Weights scale{DataType::kFLOAT, scval, len};
+    
+    float *shval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
+    for (int i = 0; i < len; i++) {
+        shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps);
+    }
+    Weights shift{DataType::kFLOAT, shval, len};
+
+    float *pval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
+    for (int i = 0; i < len; i++) {
+        pval[i] = 1.0;
+    }
+    Weights power{DataType::kFLOAT, pval, len};
+
+    weightMap[lname + ".scale"] = scale;
+    weightMap[lname + ".shift"] = shift;
+    weightMap[lname + ".power"] = power;
+    IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power);
+    assert(scale_1);
+    return scale_1;
+}
+
+ILayer* convBnLeaky(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, int ksize, int s, int g, std::string lname) {
+    Weights emptywts{DataType::kFLOAT, nullptr, 0};
+    int p = ksize / 2;
+    IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{ksize, ksize}, weightMap[lname + ".conv.weight"], emptywts);
+    assert(conv1);
+    conv1->setStrideNd(DimsHW{s, s});
+    conv1->setPaddingNd(DimsHW{p, p});
+    conv1->setNbGroups(g);
+    IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + ".bn", 1e-4);
+    auto lr = network->addActivation(*bn1->getOutput(0), ActivationType::kLEAKY_RELU);
+    lr->setAlpha(0.1);
+    return lr;
+}
+
+ILayer* focus(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int inch, int outch, int ksize, std::string lname) {
+    ISliceLayer *s1 = network->addSlice(input, Dims3{0, 0, 0}, Dims3{inch, Yolo::INPUT_H / 2, Yolo::INPUT_W / 2}, Dims3{1, 2, 2});
+    ISliceLayer *s2 = network->addSlice(input, Dims3{0, 1, 0}, Dims3{inch, Yolo::INPUT_H / 2, Yolo::INPUT_W / 2}, Dims3{1, 2, 2});
+    ISliceLayer *s3 = network->addSlice(input, Dims3{0, 0, 1}, Dims3{inch, Yolo::INPUT_H / 2, Yolo::INPUT_W / 2}, Dims3{1, 2, 2});
+    ISliceLayer *s4 = network->addSlice(input, Dims3{0, 1, 1}, Dims3{inch, Yolo::INPUT_H / 2, Yolo::INPUT_W / 2}, Dims3{1, 2, 2});
+    ITensor* inputTensors[] = {s1->getOutput(0), s2->getOutput(0), s3->getOutput(0), s4->getOutput(0)};
+    auto cat = network->addConcatenation(inputTensors, 4);
+    auto conv = convBnLeaky(network, weightMap, *cat->getOutput(0), outch, ksize, 1, 1, lname + ".conv");
+    return conv;
+}
+
+ILayer* bottleneck(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, bool shortcut, int g, float e, std::string lname) {
+    auto cv1 = convBnLeaky(network, weightMap, input, (int)((float)c2 * e), 1, 1, 1, lname + ".cv1");
+    auto cv2 = convBnLeaky(network, weightMap, *cv1->getOutput(0), c2, 3, 1, g, lname + ".cv2");
+    if (shortcut && c1 == c2) {
+        auto ew = network->addElementWise(input, *cv2->getOutput(0), ElementWiseOperation::kSUM);
+        return ew;
+    }
+    return cv2;
+}
+
+ILayer* bottleneckCSP(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, int n, bool shortcut, int g, float e, std::string lname) {
+    Weights emptywts{DataType::kFLOAT, nullptr, 0};
+    int c_ = (int)((float)c2 * e);
+    auto cv1 = convBnLeaky(network, weightMap, input, c_, 1, 1, 1, lname + ".cv1");
+    auto cv2 = network->addConvolutionNd(input, c_, DimsHW{1, 1}, weightMap[lname + ".cv2.weight"], emptywts);
+    ITensor *y1 = cv1->getOutput(0);
+    for (int i = 0; i < n; i++) {
+        auto b = bottleneck(network, weightMap, *y1, c_, c_, shortcut, g, 1.0, lname + ".m." + std::to_string(i));
+        y1 = b->getOutput(0);
+    }
+    auto cv3 = network->addConvolutionNd(*y1, c_, DimsHW{1, 1}, weightMap[lname + ".cv3.weight"], emptywts);
+
+    ITensor* inputTensors[] = {cv3->getOutput(0), cv2->getOutput(0)};
+    auto cat = network->addConcatenation(inputTensors, 2);
+
+    IScaleLayer* bn = addBatchNorm2d(network, weightMap, *cat->getOutput(0), lname + ".bn", 1e-4);
+    auto lr = network->addActivation(*bn->getOutput(0), ActivationType::kLEAKY_RELU);
+    lr->setAlpha(0.1);
+
+    auto cv4 = convBnLeaky(network, weightMap, *lr->getOutput(0), c2, 1, 1, 1, lname + ".cv4");
+    return cv4;
+}
+
+ILayer* SPP(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, int k1, int k2, int k3, std::string lname) {
+    int c_ = c1 / 2;
+    auto cv1 = convBnLeaky(network, weightMap, input, c_, 1, 1, 1, lname + ".cv1");
+
+    auto pool1 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k1, k1});
+    pool1->setPaddingNd(DimsHW{k1 / 2, k1 / 2});
+    pool1->setStrideNd(DimsHW{1, 1});
+    auto pool2 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k2, k2});
+    pool2->setPaddingNd(DimsHW{k2 / 2, k2 / 2});
+    pool2->setStrideNd(DimsHW{1, 1});
+    auto pool3 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k3, k3});
+    pool3->setPaddingNd(DimsHW{k3 / 2, k3 / 2});
+    pool3->setStrideNd(DimsHW{1, 1});
+
+    ITensor* inputTensors[] = {cv1->getOutput(0), pool1->getOutput(0), pool2->getOutput(0), pool3->getOutput(0)};
+    auto cat = network->addConcatenation(inputTensors, 4);
+
+    auto cv2 = convBnLeaky(network, weightMap, *cat->getOutput(0), c2, 1, 1, 1, lname + ".cv2");
+    return cv2;
+}
+
+int read_files_in_dir(const char *p_dir_name, std::vector<std::string> &file_names) {
+    DIR *p_dir = opendir(p_dir_name);
+    if (p_dir == nullptr) {
+        return -1;
+    }
+
+    struct dirent* p_file = nullptr;
+    while ((p_file = readdir(p_dir)) != nullptr) {
+        if (strcmp(p_file->d_name, ".") != 0 &&
+                strcmp(p_file->d_name, "..") != 0) {
+            //std::string cur_file_name(p_dir_name);
+            //cur_file_name += "/";
+            //cur_file_name += p_file->d_name;
+            std::string cur_file_name(p_file->d_name);
+            file_names.push_back(cur_file_name);
+        }
+    }
+
+    closedir(p_dir);
+    return 0;
+}
+
+#endif
+
diff --git a/yolov5/gen_wts.py b/yolov5/gen_wts.py
new file mode 100644
index 0000000..3a7857a
--- /dev/null
+++ b/yolov5/gen_wts.py
@@ -0,0 +1,18 @@
+from utils.utils import *
+import struct
+
+# Initialize
+device = torch_utils.select_device('0')
+# Load model
+model = torch.load('weights/yolov5s.pt', map_location=device)['model'].float()  # load to FP32
+model.to(device).eval()
+
+f = open('yolov5s.wts', 'w')
+f.write('{}\n'.format(len(model.state_dict().keys())))
+for k, v in model.state_dict().items():
+    vr = v.reshape(-1).cpu().numpy()
+    f.write('{} {} '.format(k, len(vr)))
+    for vv in vr:
+        f.write(' ')
+        f.write(struct.pack('>f',float(vv)).hex())
+    f.write('\n')
diff --git a/yolov5/logging.h b/yolov5/logging.h
new file mode 100644
index 0000000..602b69f
--- /dev/null
+++ b/yolov5/logging.h
@@ -0,0 +1,503 @@
+/*
+ * Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TENSORRT_LOGGING_H
+#define TENSORRT_LOGGING_H
+
+#include "NvInferRuntimeCommon.h"
+#include <cassert>
+#include <ctime>
+#include <iomanip>
+#include <iostream>
+#include <ostream>
+#include <sstream>
+#include <string>
+
+using Severity = nvinfer1::ILogger::Severity;
+
+class LogStreamConsumerBuffer : public std::stringbuf
+{
+public:
+    LogStreamConsumerBuffer(std::ostream& stream, const std::string& prefix, bool shouldLog)
+        : mOutput(stream)
+        , mPrefix(prefix)
+        , mShouldLog(shouldLog)
+    {
+    }
+
+    LogStreamConsumerBuffer(LogStreamConsumerBuffer&& other)
+        : mOutput(other.mOutput)
+    {
+    }
+
+    ~LogStreamConsumerBuffer()
+    {
+        // std::streambuf::pbase() gives a pointer to the beginning of the buffered part of the output sequence
+        // std::streambuf::pptr() gives a pointer to the current position of the output sequence
+        // if the pointer to the beginning is not equal to the pointer to the current position,
+        // call putOutput() to log the output to the stream
+        if (pbase() != pptr())
+        {
+            putOutput();
+        }
+    }
+
+    // synchronizes the stream buffer and returns 0 on success
+    // synchronizing the stream buffer consists of inserting the buffer contents into the stream,
+    // resetting the buffer and flushing the stream
+    virtual int sync()
+    {
+        putOutput();
+        return 0;
+    }
+
+    void putOutput()
+    {
+        if (mShouldLog)
+        {
+            // prepend timestamp
+            std::time_t timestamp = std::time(nullptr);
+            tm* tm_local = std::localtime(&timestamp);
+            std::cout << "[";
+            std::cout << std::setw(2) << std::setfill('0') << 1 + tm_local->tm_mon << "/";
+            std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_mday << "/";
+            std::cout << std::setw(4) << std::setfill('0') << 1900 + tm_local->tm_year << "-";
+            std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_hour << ":";
+            std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_min << ":";
+            std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_sec << "] ";
+            // std::stringbuf::str() gets the string contents of the buffer
+            // insert the buffer contents pre-appended by the appropriate prefix into the stream
+            mOutput << mPrefix << str();
+            // set the buffer to empty
+            str("");
+            // flush the stream
+            mOutput.flush();
+        }
+    }
+
+    void setShouldLog(bool shouldLog)
+    {
+        mShouldLog = shouldLog;
+    }
+
+private:
+    std::ostream& mOutput;
+    std::string mPrefix;
+    bool mShouldLog;
+};
+
+//!
+//! \class LogStreamConsumerBase
+//! \brief Convenience object used to initialize LogStreamConsumerBuffer before std::ostream in LogStreamConsumer
+//!
+class LogStreamConsumerBase
+{
+public:
+    LogStreamConsumerBase(std::ostream& stream, const std::string& prefix, bool shouldLog)
+        : mBuffer(stream, prefix, shouldLog)
+    {
+    }
+
+protected:
+    LogStreamConsumerBuffer mBuffer;
+};
+
+//!
+//! \class LogStreamConsumer
+//! \brief Convenience object used to facilitate use of C++ stream syntax when logging messages.
+//!  Order of base classes is LogStreamConsumerBase and then std::ostream.
+//!  This is because the LogStreamConsumerBase class is used to initialize the LogStreamConsumerBuffer member field
+//!  in LogStreamConsumer and then the address of the buffer is passed to std::ostream.
+//!  This is necessary to prevent the address of an uninitialized buffer from being passed to std::ostream.
+//!  Please do not change the order of the parent classes.
+//!
+class LogStreamConsumer : protected LogStreamConsumerBase, public std::ostream
+{
+public:
+    //! \brief Creates a LogStreamConsumer which logs messages with level severity.
+    //!  Reportable severity determines if the messages are severe enough to be logged.
+    LogStreamConsumer(Severity reportableSeverity, Severity severity)
+        : LogStreamConsumerBase(severityOstream(severity), severityPrefix(severity), severity <= reportableSeverity)
+        , std::ostream(&mBuffer) // links the stream buffer with the stream
+        , mShouldLog(severity <= reportableSeverity)
+        , mSeverity(severity)
+    {
+    }
+
+    LogStreamConsumer(LogStreamConsumer&& other)
+        : LogStreamConsumerBase(severityOstream(other.mSeverity), severityPrefix(other.mSeverity), other.mShouldLog)
+        , std::ostream(&mBuffer) // links the stream buffer with the stream
+        , mShouldLog(other.mShouldLog)
+        , mSeverity(other.mSeverity)
+    {
+    }
+
+    void setReportableSeverity(Severity reportableSeverity)
+    {
+        mShouldLog = mSeverity <= reportableSeverity;
+        mBuffer.setShouldLog(mShouldLog);
+    }
+
+private:
+    static std::ostream& severityOstream(Severity severity)
+    {
+        return severity >= Severity::kINFO ? std::cout : std::cerr;
+    }
+
+    static std::string severityPrefix(Severity severity)
+    {
+        switch (severity)
+        {
+        case Severity::kINTERNAL_ERROR: return "[F] ";
+        case Severity::kERROR: return "[E] ";
+        case Severity::kWARNING: return "[W] ";
+        case Severity::kINFO: return "[I] ";
+        case Severity::kVERBOSE: return "[V] ";
+        default: assert(0); return "";
+        }
+    }
+
+    bool mShouldLog;
+    Severity mSeverity;
+};
+
+//! \class Logger
+//!
+//! \brief Class which manages logging of TensorRT tools and samples
+//!
+//! \details This class provides a common interface for TensorRT tools and samples to log information to the console,
+//! and supports logging two types of messages:
+//!
+//! - Debugging messages with an associated severity (info, warning, error, or internal error/fatal)
+//! - Test pass/fail messages
+//!
+//! The advantage of having all samples use this class for logging as opposed to emitting directly to stdout/stderr is
+//! that the logic for controlling the verbosity and formatting of sample output is centralized in one location.
+//!
+//! In the future, this class could be extended to support dumping test results to a file in some standard format
+//! (for example, JUnit XML), and providing additional metadata (e.g. timing the duration of a test run).
+//!
+//! TODO: For backwards compatibility with existing samples, this class inherits directly from the nvinfer1::ILogger
+//! interface, which is problematic since there isn't a clean separation between messages coming from the TensorRT
+//! library and messages coming from the sample.
+//!
+//! In the future (once all samples are updated to use Logger::getTRTLogger() to access the ILogger) we can refactor the
+//! class to eliminate the inheritance and instead make the nvinfer1::ILogger implementation a member of the Logger
+//! object.
+
+class Logger : public nvinfer1::ILogger
+{
+public:
+    Logger(Severity severity = Severity::kWARNING)
+        : mReportableSeverity(severity)
+    {
+    }
+
+    //!
+    //! \enum TestResult
+    //! \brief Represents the state of a given test
+    //!
+    enum class TestResult
+    {
+        kRUNNING, //!< The test is running
+        kPASSED,  //!< The test passed
+        kFAILED,  //!< The test failed
+        kWAIVED   //!< The test was waived
+    };
+
+    //!
+    //! \brief Forward-compatible method for retrieving the nvinfer::ILogger associated with this Logger
+    //! \return The nvinfer1::ILogger associated with this Logger
+    //!
+    //! TODO Once all samples are updated to use this method to register the logger with TensorRT,
+    //! we can eliminate the inheritance of Logger from ILogger
+    //!
+    nvinfer1::ILogger& getTRTLogger()
+    {
+        return *this;
+    }
+
+    //!
+    //! \brief Implementation of the nvinfer1::ILogger::log() virtual method
+    //!
+    //! Note samples should not be calling this function directly; it will eventually go away once we eliminate the
+    //! inheritance from nvinfer1::ILogger
+    //!
+    void log(Severity severity, const char* msg) override
+    {
+        LogStreamConsumer(mReportableSeverity, severity) << "[TRT] " << std::string(msg) << std::endl;
+    }
+
+    //!
+    //! \brief Method for controlling the verbosity of logging output
+    //!
+    //! \param severity The logger will only emit messages that have severity of this level or higher.
+    //!
+    void setReportableSeverity(Severity severity)
+    {
+        mReportableSeverity = severity;
+    }
+
+    //!
+    //! \brief Opaque handle that holds logging information for a particular test
+    //!
+    //! This object is an opaque handle to information used by the Logger to print test results.
+    //! The sample must call Logger::defineTest() in order to obtain a TestAtom that can be used
+    //! with Logger::reportTest{Start,End}().
+    //!
+    class TestAtom
+    {
+    public:
+        TestAtom(TestAtom&&) = default;
+
+    private:
+        friend class Logger;
+
+        TestAtom(bool started, const std::string& name, const std::string& cmdline)
+            : mStarted(started)
+            , mName(name)
+            , mCmdline(cmdline)
+        {
+        }
+
+        bool mStarted;
+        std::string mName;
+        std::string mCmdline;
+    };
+
+    //!
+    //! \brief Define a test for logging
+    //!
+    //! \param[in] name The name of the test.  This should be a string starting with
+    //!                  "TensorRT" and containing dot-separated strings containing
+    //!                  the characters [A-Za-z0-9_].
+    //!                  For example, "TensorRT.sample_googlenet"
+    //! \param[in] cmdline The command line used to reproduce the test
+    //
+    //! \return a TestAtom that can be used in Logger::reportTest{Start,End}().
+    //!
+    static TestAtom defineTest(const std::string& name, const std::string& cmdline)
+    {
+        return TestAtom(false, name, cmdline);
+    }
+
+    //!
+    //! \brief A convenience overloaded version of defineTest() that accepts an array of command-line arguments
+    //!        as input
+    //!
+    //! \param[in] name The name of the test
+    //! \param[in] argc The number of command-line arguments
+    //! \param[in] argv The array of command-line arguments (given as C strings)
+    //!
+    //! \return a TestAtom that can be used in Logger::reportTest{Start,End}().
+    static TestAtom defineTest(const std::string& name, int argc, char const* const* argv)
+    {
+        auto cmdline = genCmdlineString(argc, argv);
+        return defineTest(name, cmdline);
+    }
+
+    //!
+    //! \brief Report that a test has started.
+    //!
+    //! \pre reportTestStart() has not been called yet for the given testAtom
+    //!
+    //! \param[in] testAtom The handle to the test that has started
+    //!
+    static void reportTestStart(TestAtom& testAtom)
+    {
+        reportTestResult(testAtom, TestResult::kRUNNING);
+        assert(!testAtom.mStarted);
+        testAtom.mStarted = true;
+    }
+
+    //!
+    //! \brief Report that a test has ended.
+    //!
+    //! \pre reportTestStart() has been called for the given testAtom
+    //!
+    //! \param[in] testAtom The handle to the test that has ended
+    //! \param[in] result The result of the test. Should be one of TestResult::kPASSED,
+    //!                   TestResult::kFAILED, TestResult::kWAIVED
+    //!
+    static void reportTestEnd(const TestAtom& testAtom, TestResult result)
+    {
+        assert(result != TestResult::kRUNNING);
+        assert(testAtom.mStarted);
+        reportTestResult(testAtom, result);
+    }
+
+    static int reportPass(const TestAtom& testAtom)
+    {
+        reportTestEnd(testAtom, TestResult::kPASSED);
+        return EXIT_SUCCESS;
+    }
+
+    static int reportFail(const TestAtom& testAtom)
+    {
+        reportTestEnd(testAtom, TestResult::kFAILED);
+        return EXIT_FAILURE;
+    }
+
+    static int reportWaive(const TestAtom& testAtom)
+    {
+        reportTestEnd(testAtom, TestResult::kWAIVED);
+        return EXIT_SUCCESS;
+    }
+
+    static int reportTest(const TestAtom& testAtom, bool pass)
+    {
+        return pass ? reportPass(testAtom) : reportFail(testAtom);
+    }
+
+    Severity getReportableSeverity() const
+    {
+        return mReportableSeverity;
+    }
+
+private:
+    //!
+    //! \brief returns an appropriate string for prefixing a log message with the given severity
+    //!
+    static const char* severityPrefix(Severity severity)
+    {
+        switch (severity)
+        {
+        case Severity::kINTERNAL_ERROR: return "[F] ";
+        case Severity::kERROR: return "[E] ";
+        case Severity::kWARNING: return "[W] ";
+        case Severity::kINFO: return "[I] ";
+        case Severity::kVERBOSE: return "[V] ";
+        default: assert(0); return "";
+        }
+    }
+
+    //!
+    //! \brief returns an appropriate string for prefixing a test result message with the given result
+    //!
+    static const char* testResultString(TestResult result)
+    {
+        switch (result)
+        {
+        case TestResult::kRUNNING: return "RUNNING";
+        case TestResult::kPASSED: return "PASSED";
+        case TestResult::kFAILED: return "FAILED";
+        case TestResult::kWAIVED: return "WAIVED";
+        default: assert(0); return "";
+        }
+    }
+
+    //!
+    //! \brief returns an appropriate output stream (cout or cerr) to use with the given severity
+    //!
+    static std::ostream& severityOstream(Severity severity)
+    {
+        return severity >= Severity::kINFO ? std::cout : std::cerr;
+    }
+
+    //!
+    //! \brief method that implements logging test results
+    //!
+    static void reportTestResult(const TestAtom& testAtom, TestResult result)
+    {
+        severityOstream(Severity::kINFO) << "&&&& " << testResultString(result) << " " << testAtom.mName << " # "
+                                         << testAtom.mCmdline << std::endl;
+    }
+
+    //!
+    //! \brief generate a command line string from the given (argc, argv) values
+    //!
+    static std::string genCmdlineString(int argc, char const* const* argv)
+    {
+        std::stringstream ss;
+        for (int i = 0; i < argc; i++)
+        {
+            if (i > 0)
+                ss << " ";
+            ss << argv[i];
+        }
+        return ss.str();
+    }
+
+    Severity mReportableSeverity;
+};
+
+namespace
+{
+
+//!
+//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kVERBOSE
+//!
+//! Example usage:
+//!
+//!     LOG_VERBOSE(logger) << "hello world" << std::endl;
+//!
+inline LogStreamConsumer LOG_VERBOSE(const Logger& logger)
+{
+    return LogStreamConsumer(logger.getReportableSeverity(), Severity::kVERBOSE);
+}
+
+//!
+//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINFO
+//!
+//! Example usage:
+//!
+//!     LOG_INFO(logger) << "hello world" << std::endl;
+//!
+inline LogStreamConsumer LOG_INFO(const Logger& logger)
+{
+    return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINFO);
+}
+
+//!
+//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kWARNING
+//!
+//! Example usage:
+//!
+//!     LOG_WARN(logger) << "hello world" << std::endl;
+//!
+inline LogStreamConsumer LOG_WARN(const Logger& logger)
+{
+    return LogStreamConsumer(logger.getReportableSeverity(), Severity::kWARNING);
+}
+
+//!
+//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kERROR
+//!
+//! Example usage:
+//!
+//!     LOG_ERROR(logger) << "hello world" << std::endl;
+//!
+inline LogStreamConsumer LOG_ERROR(const Logger& logger)
+{
+    return LogStreamConsumer(logger.getReportableSeverity(), Severity::kERROR);
+}
+
+//!
+//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINTERNAL_ERROR
+//         ("fatal" severity)
+//!
+//! Example usage:
+//!
+//!     LOG_FATAL(logger) << "hello world" << std::endl;
+//!
+inline LogStreamConsumer LOG_FATAL(const Logger& logger)
+{
+    return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINTERNAL_ERROR);
+}
+
+} // anonymous namespace
+
+#endif // TENSORRT_LOGGING_H
diff --git a/yolov5/utils.h b/yolov5/utils.h
new file mode 100644
index 0000000..0de663c
--- /dev/null
+++ b/yolov5/utils.h
@@ -0,0 +1,94 @@
+#ifndef __TRT_UTILS_H_
+#define __TRT_UTILS_H_
+
+#include <iostream>
+#include <vector>
+#include <algorithm>
+#include <cudnn.h>
+
+#ifndef CUDA_CHECK
+
+#define CUDA_CHECK(callstr)                                                                    \
+    {                                                                                          \
+        cudaError_t error_code = callstr;                                                      \
+        if (error_code != cudaSuccess) {                                                       \
+            std::cerr << "CUDA error " << error_code << " at " << __FILE__ << ":" << __LINE__; \
+            assert(0);                                                                         \
+        }                                                                                      \
+    }
+
+#endif
+
+namespace Tn
+{
+    class Profiler : public nvinfer1::IProfiler
+    {
+    public:
+        void printLayerTimes(int itrationsTimes)
+        {
+            float totalTime = 0;
+            for (size_t i = 0; i < mProfile.size(); i++)
+            {
+                printf("%-40.40s %4.3fms\n", mProfile[i].first.c_str(), mProfile[i].second / itrationsTimes);
+                totalTime += mProfile[i].second;
+            }
+            printf("Time over all layers: %4.3f\n", totalTime / itrationsTimes);
+        }
+    private:
+        typedef std::pair<std::string, float> Record;
+        std::vector<Record> mProfile;
+
+        virtual void reportLayerTime(const char* layerName, float ms)
+        {
+            auto record = std::find_if(mProfile.begin(), mProfile.end(), [&](const Record& r){ return r.first == layerName; });
+            if (record == mProfile.end())
+                mProfile.push_back(std::make_pair(layerName, ms));
+            else
+                record->second += ms;
+        }
+    };
+
+    //Logger for TensorRT info/warning/errors
+    class Logger : public nvinfer1::ILogger
+    {
+    public:
+
+        Logger(): Logger(Severity::kWARNING) {}
+
+        Logger(Severity severity): reportableSeverity(severity) {}
+
+        void log(Severity severity, const char* msg) override
+        {
+            // suppress messages with severity enum value greater than the reportable
+            if (severity > reportableSeverity) return;
+
+            switch (severity)
+            {
+                case Severity::kINTERNAL_ERROR: std::cerr << "INTERNAL_ERROR: "; break;
+                case Severity::kERROR: std::cerr << "ERROR: "; break;
+                case Severity::kWARNING: std::cerr << "WARNING: "; break;
+                case Severity::kINFO: std::cerr << "INFO: "; break;
+                default: std::cerr << "UNKNOWN: "; break;
+            }
+            std::cerr << msg << std::endl;
+        }
+
+        Severity reportableSeverity{Severity::kWARNING};
+    };
+
+    template<typename T> 
+    void write(char*& buffer, const T& val)
+    {
+        *reinterpret_cast<T*>(buffer) = val;
+        buffer += sizeof(T);
+    }
+
+    template<typename T> 
+    void read(const char*& buffer, T& val)
+    {
+        val = *reinterpret_cast<const T*>(buffer);
+        buffer += sizeof(T);
+    }
+}
+
+#endif
\ No newline at end of file
diff --git a/yolov5/yololayer.cu b/yolov5/yololayer.cu
new file mode 100644
index 0000000..aef5d60
--- /dev/null
+++ b/yolov5/yololayer.cu
@@ -0,0 +1,261 @@
+#include <assert.h>
+#include "yololayer.h"
+#include "utils.h"
+
+using namespace Yolo;
+
+namespace nvinfer1
+{
+    YoloLayerPlugin::YoloLayerPlugin()
+    {
+        mClassCount = CLASS_NUM;
+        mYoloKernel.clear();
+        mYoloKernel.push_back(yolo1);
+        mYoloKernel.push_back(yolo2);
+        mYoloKernel.push_back(yolo3);
+
+        mKernelCount = mYoloKernel.size();
+    }
+    
+    YoloLayerPlugin::~YoloLayerPlugin()
+    {
+    }
+
+    // create the plugin at runtime from a byte stream
+    YoloLayerPlugin::YoloLayerPlugin(const void* data, size_t length)
+    {
+        using namespace Tn;
+        const char *d = reinterpret_cast<const char *>(data), *a = d;
+        read(d, mClassCount);
+        read(d, mThreadCount);
+        read(d, mKernelCount);
+        mYoloKernel.resize(mKernelCount);
+        auto kernelSize = mKernelCount*sizeof(YoloKernel);
+        memcpy(mYoloKernel.data(),d,kernelSize);
+        d += kernelSize;
+
+        assert(d == a + length);
+    }
+
+    void YoloLayerPlugin::serialize(void* buffer) const
+    {
+        using namespace Tn;
+        char* d = static_cast<char*>(buffer), *a = d;
+        write(d, mClassCount);
+        write(d, mThreadCount);
+        write(d, mKernelCount);
+        auto kernelSize = mKernelCount*sizeof(YoloKernel);
+        memcpy(d,mYoloKernel.data(),kernelSize);
+        d += kernelSize;
+
+        assert(d == a + getSerializationSize());
+    }
+    
+    size_t YoloLayerPlugin::getSerializationSize() const
+    {  
+        return sizeof(mClassCount) + sizeof(mThreadCount) + sizeof(mKernelCount)  + sizeof(Yolo::YoloKernel) * mYoloKernel.size();
+    }
+
+    int YoloLayerPlugin::initialize()
+    { 
+        return 0;
+    }
+    
+    Dims YoloLayerPlugin::getOutputDimensions(int index, const Dims* inputs, int nbInputDims)
+    {
+        //output the result to channel
+        int totalsize = MAX_OUTPUT_BBOX_COUNT * sizeof(Detection) / sizeof(float);
+
+        return Dims3(totalsize + 1, 1, 1);
+    }
+
+    // Set plugin namespace
+    void YoloLayerPlugin::setPluginNamespace(const char* pluginNamespace)
+    {
+        mPluginNamespace = pluginNamespace;
+    }
+
+    const char* YoloLayerPlugin::getPluginNamespace() const
+    {
+        return mPluginNamespace;
+    }
+
+    // Return the DataType of the plugin output at the requested index
+    DataType YoloLayerPlugin::getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const
+    {
+        return DataType::kFLOAT;
+    }
+
+    // Return true if output tensor is broadcast across a batch.
+    bool YoloLayerPlugin::isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const
+    {
+        return false;
+    }
+
+    // Return true if plugin can use input that is broadcast across batch without replication.
+    bool YoloLayerPlugin::canBroadcastInputAcrossBatch(int inputIndex) const
+    {
+        return false;
+    }
+
+    void YoloLayerPlugin::configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput)
+    {
+    }
+
+    // Attach the plugin object to an execution context and grant the plugin the access to some context resource.
+    void YoloLayerPlugin::attachToContext(cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator)
+    {
+    }
+
+    // Detach the plugin object from its execution context.
+    void YoloLayerPlugin::detachFromContext() {}
+
+    const char* YoloLayerPlugin::getPluginType() const
+    {
+        return "YoloLayer_TRT";
+    }
+
+    const char* YoloLayerPlugin::getPluginVersion() const
+    {
+        return "1";
+    }
+
+    void YoloLayerPlugin::destroy()
+    {
+        delete this;
+    }
+
+    // Clone the plugin
+    IPluginV2IOExt* YoloLayerPlugin::clone() const
+    {
+        YoloLayerPlugin *p = new YoloLayerPlugin();
+        p->setPluginNamespace(mPluginNamespace);
+        return p;
+    }
+
+    __device__ float Logist(float data){ return 1.0f / (1.0f + expf(-data)); };
+
+    __global__ void CalDetection(const float *input, float *output,int noElements, 
+            int yoloWidth,int yoloHeight,const float anchors[CHECK_COUNT*2],int classes,int outputElem) {
+ 
+        int idx = threadIdx.x + blockDim.x * blockIdx.x;
+        if (idx >= noElements) return;
+
+        int total_grid = yoloWidth * yoloHeight;
+        int bnIdx = idx / total_grid;
+        idx = idx - total_grid*bnIdx;
+        int info_len_i = 5 + classes;
+        const float* curInput = input + bnIdx * (info_len_i * total_grid * CHECK_COUNT);
+
+        for (int k = 0; k < 3; ++k) {
+            float box_prob = Logist(curInput[idx + k * info_len_i * total_grid + 4 * total_grid]);
+            if (box_prob < IGNORE_THRESH) continue;
+            int class_id = 0;
+            float max_cls_prob = 0.0;
+            for (int i = 5; i < info_len_i; ++i) {
+                float p = Logist(curInput[idx + k * info_len_i * total_grid + i * total_grid]);
+                if (p > max_cls_prob) {
+                    max_cls_prob = p;
+                    class_id = i - 5;
+                }
+            }
+            float *res_count = output + bnIdx*outputElem;
+            int count = (int)atomicAdd(res_count, 1);
+            if (count >= MAX_OUTPUT_BBOX_COUNT) return;
+            char* data = (char * )res_count + sizeof(float) + count*sizeof(Detection);
+            Detection* det =  (Detection*)(data);
+
+            int row = idx / yoloWidth;
+            int col = idx % yoloWidth;
+
+            //Location
+            det->bbox[0] = (col - 0.5f + 2.0f * Logist(curInput[idx + k * info_len_i * total_grid + 0 * total_grid])) * INPUT_W / yoloWidth;
+            det->bbox[1] = (row - 0.5f + 2.0f * Logist(curInput[idx + k * info_len_i * total_grid + 1 * total_grid])) * INPUT_H / yoloHeight;
+            det->bbox[2] = 2.0f * Logist(curInput[idx + k * info_len_i * total_grid + 2 * total_grid]);
+            det->bbox[2] = det->bbox[2] * det->bbox[2] * anchors[2*k];
+            det->bbox[3] = 2.0f * Logist(curInput[idx + k * info_len_i * total_grid + 3 * total_grid]);
+            det->bbox[3] = det->bbox[3] * det->bbox[3] * anchors[2*k + 1];
+            det->conf = box_prob * max_cls_prob;
+            det->class_id = class_id;
+        }
+    }
+
+    void YoloLayerPlugin::forwardGpu(const float *const * inputs, float* output, cudaStream_t stream, int batchSize) {
+        void* devAnchor;
+        size_t AnchorLen = sizeof(float)* CHECK_COUNT*2;
+        CUDA_CHECK(cudaMalloc(&devAnchor,AnchorLen));
+
+        int outputElem = 1 + MAX_OUTPUT_BBOX_COUNT * sizeof(Detection) / sizeof(float);
+
+        for(int idx = 0 ; idx < batchSize; ++idx) {
+            CUDA_CHECK(cudaMemset(output + idx*outputElem, 0, sizeof(float)));
+        }
+        int numElem = 0;
+        for (unsigned int i = 0;i< mYoloKernel.size();++i)
+        {
+            const auto& yolo = mYoloKernel[i];
+            numElem = yolo.width*yolo.height*batchSize;
+            if (numElem < mThreadCount)
+                mThreadCount = numElem;
+            CUDA_CHECK(cudaMemcpy(devAnchor, yolo.anchors, AnchorLen, cudaMemcpyHostToDevice));
+            CalDetection<<< (yolo.width*yolo.height*batchSize + mThreadCount - 1) / mThreadCount, mThreadCount>>>
+                (inputs[i],output, numElem, yolo.width, yolo.height, (float *)devAnchor, mClassCount ,outputElem);
+        }
+
+        CUDA_CHECK(cudaFree(devAnchor));
+    }
+
+
+    int YoloLayerPlugin::enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream)
+    {
+        //assert(batchSize == 1);
+        //GPU
+        //CUDA_CHECK(cudaStreamSynchronize(stream));
+        forwardGpu((const float *const *)inputs, (float*)outputs[0], stream, batchSize);
+
+        return 0;
+    }
+
+    PluginFieldCollection YoloPluginCreator::mFC{};
+    std::vector<PluginField> YoloPluginCreator::mPluginAttributes;
+
+    YoloPluginCreator::YoloPluginCreator()
+    {
+        mPluginAttributes.clear();
+
+        mFC.nbFields = mPluginAttributes.size();
+        mFC.fields = mPluginAttributes.data();
+    }
+
+    const char* YoloPluginCreator::getPluginName() const
+    {
+            return "YoloLayer_TRT";
+    }
+
+    const char* YoloPluginCreator::getPluginVersion() const
+    {
+            return "1";
+    }
+
+    const PluginFieldCollection* YoloPluginCreator::getFieldNames()
+    {
+            return &mFC;
+    }
+
+    IPluginV2IOExt* YoloPluginCreator::createPlugin(const char* name, const PluginFieldCollection* fc)
+    {
+        YoloLayerPlugin* obj = new YoloLayerPlugin();
+        obj->setPluginNamespace(mNamespace.c_str());
+        return obj;
+    }
+
+    IPluginV2IOExt* YoloPluginCreator::deserializePlugin(const char* name, const void* serialData, size_t serialLength)
+    {
+        // This object will be deleted when the network is destroyed, which will
+        // call MishPlugin::destroy()
+        YoloLayerPlugin* obj = new YoloLayerPlugin(serialData, serialLength);
+        obj->setPluginNamespace(mNamespace.c_str());
+        return obj;
+    }
+
+}
diff --git a/yolov5/yololayer.h b/yolov5/yololayer.h
new file mode 100644
index 0000000..cc642af
--- /dev/null
+++ b/yolov5/yololayer.h
@@ -0,0 +1,153 @@
+#ifndef _YOLO_LAYER_H
+#define _YOLO_LAYER_H
+
+#include <vector>
+#include <string>
+#include "NvInfer.h"
+
+namespace Yolo
+{
+    static constexpr int CHECK_COUNT = 3;
+    static constexpr float IGNORE_THRESH = 0.1f;
+    static constexpr int MAX_OUTPUT_BBOX_COUNT = 1000;
+    static constexpr int CLASS_NUM = 80;
+    static constexpr int INPUT_H = 608;
+    static constexpr int INPUT_W = 608;
+
+    struct YoloKernel
+    {
+        int width;
+        int height;
+        float anchors[CHECK_COUNT*2];
+    };
+
+    static constexpr YoloKernel yolo1 = {
+        INPUT_W / 32,
+        INPUT_H / 32,
+        {116,90,  156,198,  373,326}
+    };
+    static constexpr YoloKernel yolo2 = {
+        INPUT_W / 16,
+        INPUT_H / 16,
+        {30,61,  62,45,  59,119}
+    };
+    static constexpr YoloKernel yolo3 = {
+        INPUT_W / 8,
+        INPUT_H / 8,
+        {10,13,  16,30,  33,23}
+    };
+
+    static constexpr int LOCATIONS = 4;
+    struct alignas(float) Detection{
+        //center_x center_y w h
+        float bbox[LOCATIONS];
+        float conf;  // bbox_conf * cls_conf
+        float class_id;
+    };
+}
+
+namespace nvinfer1
+{
+    class YoloLayerPlugin: public IPluginV2IOExt
+    {
+        public:
+            explicit YoloLayerPlugin();
+            YoloLayerPlugin(const void* data, size_t length);
+
+            ~YoloLayerPlugin();
+
+            int getNbOutputs() const override
+            {
+                return 1;
+            }
+
+            Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override;
+
+            int initialize() override;
+
+            virtual void terminate() override {};
+
+            virtual size_t getWorkspaceSize(int maxBatchSize) const override { return 0;}
+
+            virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override;
+
+            virtual size_t getSerializationSize() const override;
+
+            virtual void serialize(void* buffer) const override;
+
+            bool supportsFormatCombination(int pos, const PluginTensorDesc* inOut, int nbInputs, int nbOutputs) const override {
+                return inOut[pos].format == TensorFormat::kLINEAR && inOut[pos].type == DataType::kFLOAT;
+            }
+
+            const char* getPluginType() const override;
+
+            const char* getPluginVersion() const override;
+
+            void destroy() override;
+
+            IPluginV2IOExt* clone() const override;
+
+            void setPluginNamespace(const char* pluginNamespace) override;
+
+            const char* getPluginNamespace() const override;
+
+            DataType getOutputDataType(int index, const nvinfer1::DataType* inputTypes, int nbInputs) const override;
+
+            bool isOutputBroadcastAcrossBatch(int outputIndex, const bool* inputIsBroadcasted, int nbInputs) const override;
+
+            bool canBroadcastInputAcrossBatch(int inputIndex) const override;
+
+            void attachToContext(
+                    cudnnContext* cudnnContext, cublasContext* cublasContext, IGpuAllocator* gpuAllocator) override;
+
+            void configurePlugin(const PluginTensorDesc* in, int nbInput, const PluginTensorDesc* out, int nbOutput) override;
+
+            void detachFromContext() override;
+
+        private:
+            void forwardGpu(const float *const * inputs,float * output, cudaStream_t stream,int batchSize = 1);
+            int mClassCount;
+            int mKernelCount;
+            std::vector<Yolo::YoloKernel> mYoloKernel;
+            int mThreadCount = 256;
+            const char* mPluginNamespace;
+    };
+
+    class YoloPluginCreator : public IPluginCreator
+    {
+        public:
+            YoloPluginCreator();
+
+            ~YoloPluginCreator() override = default;
+
+            const char* getPluginName() const override;
+
+            const char* getPluginVersion() const override;
+
+            const PluginFieldCollection* getFieldNames() override;
+
+            IPluginV2IOExt* createPlugin(const char* name, const PluginFieldCollection* fc) override;
+
+            IPluginV2IOExt* deserializePlugin(const char* name, const void* serialData, size_t serialLength) override;
+
+            void setPluginNamespace(const char* libNamespace) override
+            {
+                mNamespace = libNamespace;
+            }
+
+            const char* getPluginNamespace() const override
+            {
+                return mNamespace.c_str();
+            }
+
+        private:
+            std::string mNamespace;
+            static PluginFieldCollection mFC;
+            static std::vector<PluginField> mPluginAttributes;
+    };
+
+
+
+};
+
+#endif 
diff --git a/yolov5/yolov5s.cpp b/yolov5/yolov5s.cpp
index 7a115f2..c9dbd48 100644
--- a/yolov5/yolov5s.cpp
+++ b/yolov5/yolov5s.cpp
@@ -1,289 +1,23 @@
-#include <fstream>
 #include <iostream>
-#include <map>
-#include <sstream>
-#include <vector>
 #include <chrono>
-#include <opencv2/opencv.hpp>
-#include <dirent.h>
-#include "NvInfer.h"
 #include "cuda_runtime_api.h"
 #include "logging.h"
-#include "yololayer.h"
-
-#define CHECK(status) \
-    do\
-    {\
-        auto ret = (status);\
-        if (ret != 0)\
-        {\
-            std::cerr << "Cuda failure: " << ret << std::endl;\
-            abort();\
-        }\
-    } while (0)
-
+#include "common.hpp"
 
 #define USE_FP16  // comment out this if want to use FP32
 #define DEVICE 0  // GPU id
 #define NMS_THRESH 0.5
-#define BBOX_CONF_THRESH 0.4
-
-using namespace nvinfer1;
+#define CONF_THRESH 0.4
 
 // stuff we know about the network and the input/output blobs
 static const int INPUT_H = Yolo::INPUT_H;
 static const int INPUT_W = Yolo::INPUT_W;
-static const int OUTPUT_SIZE = 1000 * 7 + 1;  // we assume the yololayer outputs no more than 1000 boxes that conf >= 0.1
+static const int OUTPUT_SIZE = Yolo::MAX_OUTPUT_BBOX_COUNT * sizeof(Yolo::Detection) / sizeof(float) + 1;  // we assume the yololayer outputs no more than 1000 boxes that conf >= 0.1
 const char* INPUT_BLOB_NAME = "data";
 const char* OUTPUT_BLOB_NAME = "prob";
 static Logger gLogger;
 REGISTER_TENSORRT_PLUGIN(YoloPluginCreator);
 
-cv::Mat preprocess_img(cv::Mat& img) {
-    int w, h, x, y;
-    float r_w = INPUT_W / (img.cols*1.0);
-    float r_h = INPUT_H / (img.rows*1.0);
-    if (r_h > r_w) {
-        w = INPUT_W;
-        h = r_w * img.rows;
-        x = 0;
-        y = (INPUT_H - h) / 2;
-    } else {
-        w = r_h* img.cols;
-        h = INPUT_H;
-        x = (INPUT_W - w) / 2;
-        y = 0;
-    }
-    cv::Mat re(h, w, CV_8UC3);
-    cv::resize(img, re, re.size(), 0, 0, cv::INTER_CUBIC);
-    cv::Mat out(INPUT_H, INPUT_W, CV_8UC3, cv::Scalar(128, 128, 128));
-    re.copyTo(out(cv::Rect(x, y, re.cols, re.rows)));
-    return out;
-}
-
-cv::Rect get_rect(cv::Mat& img, float bbox[4]) {
-    int l, r, t, b;
-    float r_w = INPUT_W / (img.cols * 1.0);
-    float r_h = INPUT_H / (img.rows * 1.0);
-    if (r_h > r_w) {
-        l = bbox[0] - bbox[2]/2.f;
-        r = bbox[0] + bbox[2]/2.f;
-        t = bbox[1] - bbox[3]/2.f - (INPUT_H - r_w * img.rows) / 2;
-        b = bbox[1] + bbox[3]/2.f - (INPUT_H - r_w * img.rows) / 2;
-        l = l / r_w;
-        r = r / r_w;
-        t = t / r_w;
-        b = b / r_w;
-    } else {
-        l = bbox[0] - bbox[2]/2.f - (INPUT_W - r_h * img.cols) / 2;
-        r = bbox[0] + bbox[2]/2.f - (INPUT_W - r_h * img.cols) / 2;
-        t = bbox[1] - bbox[3]/2.f;
-        b = bbox[1] + bbox[3]/2.f;
-        l = l / r_h;
-        r = r / r_h;
-        t = t / r_h;
-        b = b / r_h;
-    }
-    return cv::Rect(l, t, r-l, b-t);
-}
-
-float iou(float lbox[4], float rbox[4]) {
-    float interBox[] = {
-        std::max(lbox[0] - lbox[2]/2.f , rbox[0] - rbox[2]/2.f), //left
-        std::min(lbox[0] + lbox[2]/2.f , rbox[0] + rbox[2]/2.f), //right
-        std::max(lbox[1] - lbox[3]/2.f , rbox[1] - rbox[3]/2.f), //top
-        std::min(lbox[1] + lbox[3]/2.f , rbox[1] + rbox[3]/2.f), //bottom
-    };
-
-    if(interBox[2] > interBox[3] || interBox[0] > interBox[1])
-        return 0.0f;
-
-    float interBoxS =(interBox[1]-interBox[0])*(interBox[3]-interBox[2]);
-    return interBoxS/(lbox[2]*lbox[3] + rbox[2]*rbox[3] -interBoxS);
-}
-
-bool cmp(Yolo::Detection& a, Yolo::Detection& b) {
-    return a.det_confidence * a.class_confidence > b.det_confidence * b.class_confidence;
-}
-
-void nms(std::vector<Yolo::Detection>& res, float *output, float nms_thresh = NMS_THRESH) {
-    std::map<float, std::vector<Yolo::Detection>> m;
-    for (int i = 0; i < output[0] && i < 1000; i++) {
-        if (output[1 + 7 * i + 4] * output[1 + 7 * i + 6] <= BBOX_CONF_THRESH) continue;
-        Yolo::Detection det;
-        memcpy(&det, &output[1 + 7 * i], 7 * sizeof(float));
-        if (m.count(det.class_id) == 0) m.emplace(det.class_id, std::vector<Yolo::Detection>());
-        m[det.class_id].push_back(det);
-    }
-    for (auto it = m.begin(); it != m.end(); it++) {
-        //std::cout << it->second[0].class_id << " --- " << std::endl;
-        auto& dets = it->second;
-        std::sort(dets.begin(), dets.end(), cmp);
-        for (size_t m = 0; m < dets.size(); ++m) {
-            auto& item = dets[m];
-            res.push_back(item);
-            for (size_t n = m + 1; n < dets.size(); ++n) {
-                if (iou(item.bbox, dets[n].bbox) > nms_thresh) {
-                    dets.erase(dets.begin()+n);
-                    --n;
-                }
-            }
-        }
-    }
-}
-
-// TensorRT weight files have a simple space delimited format:
-// [type] [size] <data x size in hex>
-std::map<std::string, Weights> loadWeights(const std::string file) {
-    std::cout << "Loading weights: " << file << std::endl;
-    std::map<std::string, Weights> weightMap;
-
-    // Open weights file
-    std::ifstream input(file);
-    assert(input.is_open() && "Unable to load weight file.");
-
-    // Read number of weight blobs
-    int32_t count;
-    input >> count;
-    assert(count > 0 && "Invalid weight map file.");
-
-    while (count--)
-    {
-        Weights wt{DataType::kFLOAT, nullptr, 0};
-        uint32_t size;
-
-        // Read name and type of blob
-        std::string name;
-        input >> name >> std::dec >> size;
-        wt.type = DataType::kFLOAT;
-
-        // Load blob
-        uint32_t* val = reinterpret_cast<uint32_t*>(malloc(sizeof(val) * size));
-        for (uint32_t x = 0, y = size; x < y; ++x)
-        {
-            input >> std::hex >> val[x];
-        }
-        wt.values = val;
-        
-        wt.count = size;
-        weightMap[name] = wt;
-    }
-
-    return weightMap;
-}
-
-IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, std::string lname, float eps) {
-    float *gamma = (float*)weightMap[lname + ".weight"].values;
-    float *beta = (float*)weightMap[lname + ".bias"].values;
-    float *mean = (float*)weightMap[lname + ".running_mean"].values;
-    float *var = (float*)weightMap[lname + ".running_var"].values;
-    int len = weightMap[lname + ".running_var"].count;
-
-    float *scval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
-    for (int i = 0; i < len; i++) {
-        scval[i] = gamma[i] / sqrt(var[i] + eps);
-    }
-    Weights scale{DataType::kFLOAT, scval, len};
-    
-    float *shval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
-    for (int i = 0; i < len; i++) {
-        shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps);
-    }
-    Weights shift{DataType::kFLOAT, shval, len};
-
-    float *pval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
-    for (int i = 0; i < len; i++) {
-        pval[i] = 1.0;
-    }
-    Weights power{DataType::kFLOAT, pval, len};
-
-    weightMap[lname + ".scale"] = scale;
-    weightMap[lname + ".shift"] = shift;
-    weightMap[lname + ".power"] = power;
-    IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power);
-    assert(scale_1);
-    return scale_1;
-}
-
-ILayer* convBnLeaky(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int outch, int ksize, int s, int g, std::string lname) {
-    Weights emptywts{DataType::kFLOAT, nullptr, 0};
-    int p = ksize / 2;
-    IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{ksize, ksize}, weightMap[lname + ".conv.weight"], emptywts);
-    assert(conv1);
-    conv1->setStrideNd(DimsHW{s, s});
-    conv1->setPaddingNd(DimsHW{p, p});
-    conv1->setNbGroups(g);
-    IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), lname + ".bn", 1e-4);
-    auto lr = network->addActivation(*bn1->getOutput(0), ActivationType::kLEAKY_RELU);
-    lr->setAlpha(0.1);
-    return lr;
-}
-
-ILayer* focus(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int inch, int outch, int ksize, std::string lname) {
-    ISliceLayer *s1 = network->addSlice(input, Dims3{0, 0, 0}, Dims3{inch, INPUT_H / 2, INPUT_W / 2}, Dims3{1, 2, 2});
-    ISliceLayer *s2 = network->addSlice(input, Dims3{0, 1, 0}, Dims3{inch, INPUT_H / 2, INPUT_W / 2}, Dims3{1, 2, 2});
-    ISliceLayer *s3 = network->addSlice(input, Dims3{0, 0, 1}, Dims3{inch, INPUT_H / 2, INPUT_W / 2}, Dims3{1, 2, 2});
-    ISliceLayer *s4 = network->addSlice(input, Dims3{0, 1, 1}, Dims3{inch, INPUT_H / 2, INPUT_W / 2}, Dims3{1, 2, 2});
-    ITensor* inputTensors[] = {s1->getOutput(0), s2->getOutput(0), s3->getOutput(0), s4->getOutput(0)};
-    auto cat = network->addConcatenation(inputTensors, 4);
-    auto conv = convBnLeaky(network, weightMap, *cat->getOutput(0), outch, ksize, 1, 1, lname + ".conv");
-    return conv;
-}
-
-ILayer* bottleneck(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, bool shortcut, int g, float e, std::string lname) {
-    auto cv1 = convBnLeaky(network, weightMap, input, (int)((float)c2 * e), 1, 1, 1, lname + ".cv1");
-    auto cv2 = convBnLeaky(network, weightMap, *cv1->getOutput(0), c2, 3, 1, g, lname + ".cv2");
-    if (shortcut && c1 == c2) {
-        auto ew = network->addElementWise(input, *cv2->getOutput(0), ElementWiseOperation::kSUM);
-        return ew;
-    }
-    return cv2;
-}
-
-ILayer* bottleneckCSP(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, int n, bool shortcut, int g, float e, std::string lname) {
-    Weights emptywts{DataType::kFLOAT, nullptr, 0};
-    int c_ = (int)((float)c2 * e);
-    auto cv1 = convBnLeaky(network, weightMap, input, c_, 1, 1, 1, lname + ".cv1");
-    auto cv2 = network->addConvolutionNd(input, c_, DimsHW{1, 1}, weightMap[lname + ".cv2.weight"], emptywts);
-    ITensor *y1 = cv1->getOutput(0);
-    for (int i = 0; i < n; i++) {
-        auto b = bottleneck(network, weightMap, *y1, c_, c_, shortcut, g, 1.0, lname + ".m." + std::to_string(i));
-        y1 = b->getOutput(0);
-    }
-    auto cv3 = network->addConvolutionNd(*y1, c_, DimsHW{1, 1}, weightMap[lname + ".cv3.weight"], emptywts);
-
-    ITensor* inputTensors[] = {cv3->getOutput(0), cv2->getOutput(0)};
-    auto cat = network->addConcatenation(inputTensors, 2);
-
-    IScaleLayer* bn = addBatchNorm2d(network, weightMap, *cat->getOutput(0), lname + ".bn", 1e-4);
-    auto lr = network->addActivation(*bn->getOutput(0), ActivationType::kLEAKY_RELU);
-    lr->setAlpha(0.1);
-
-    auto cv4 = convBnLeaky(network, weightMap, *lr->getOutput(0), c2, 1, 1, 1, lname + ".cv4");
-    return cv4;
-}
-
-ILayer* SPP(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input, int c1, int c2, int k1, int k2, int k3, std::string lname) {
-    int c_ = c1 / 2;
-    auto cv1 = convBnLeaky(network, weightMap, input, c_, 1, 1, 1, lname + ".cv1");
-
-    auto pool1 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k1, k1});
-    pool1->setPaddingNd(DimsHW{k1 / 2, k1 / 2});
-    pool1->setStrideNd(DimsHW{1, 1});
-    auto pool2 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k2, k2});
-    pool2->setPaddingNd(DimsHW{k2 / 2, k2 / 2});
-    pool2->setStrideNd(DimsHW{1, 1});
-    auto pool3 = network->addPoolingNd(*cv1->getOutput(0), PoolingType::kMAX, DimsHW{k3, k3});
-    pool3->setPaddingNd(DimsHW{k3 / 2, k3 / 2});
-    pool3->setStrideNd(DimsHW{1, 1});
-
-    ITensor* inputTensors[] = {cv1->getOutput(0), pool1->getOutput(0), pool2->getOutput(0), pool3->getOutput(0)};
-    auto cat = network->addConcatenation(inputTensors, 4);
-
-    auto cv2 = convBnLeaky(network, weightMap, *cat->getOutput(0), c2, 1, 1, 1, lname + ".cv2");
-    return cv2;
-}
-
 // Creat the engine using only the API and not any parser.
 ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig* config, DataType dt) {
     INetworkDefinition* network = builder->createNetworkV2(0U);
@@ -417,28 +151,6 @@ void doInference(IExecutionContext& context, float* input, float* output, int ba
     CHECK(cudaFree(buffers[outputIndex]));
 }
 
-int read_files_in_dir(const char *p_dir_name, std::vector<std::string> &file_names) {
-    DIR *p_dir = opendir(p_dir_name);
-    if (p_dir == nullptr) {
-        return -1;
-    }
-
-    struct dirent* p_file = nullptr;
-    while ((p_file = readdir(p_dir)) != nullptr) {
-        if (strcmp(p_file->d_name, ".") != 0 &&
-                strcmp(p_file->d_name, "..") != 0) {
-            //std::string cur_file_name(p_dir_name);
-            //cur_file_name += "/";
-            //cur_file_name += p_file->d_name;
-            std::string cur_file_name(p_file->d_name);
-            file_names.push_back(cur_file_name);
-        }
-    }
-
-    closedir(p_dir);
-    return 0;
-}
-
 int main(int argc, char** argv) {
     cudaSetDevice(DEVICE);
     // create a model using the API directly and serialize it to a stream
@@ -513,7 +225,7 @@ int main(int argc, char** argv) {
         auto end = std::chrono::system_clock::now();
         std::cout << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms" << std::endl;
         std::vector<Yolo::Detection> res;
-        nms(res, prob);
+        nms(res, prob, CONF_THRESH, NMS_THRESH);
         for (int i=0; i<20; i++) {
             std::cout << prob[i] << ",";
         }