diff --git a/README.md b/README.md
index 0c2dea7..198e58a 100644
--- a/README.md
+++ b/README.md
@@ -78,6 +78,7 @@ Some tricky operations encountered in these models, already solved, but might ha
 |-|-|:-:|:-:|:-:|:-:|
 | YOLOv3(darknet53) | Xavier | 1 | FP16 | 320x320 | 55 |
 | YOLOv3-spp(darknet53) | Xeon E5-2620/GTX1080 | 1 | FP32 | 256x416 | 94 |
+| YOLOv3-spp(darknet53) | Xeon E5-2620/GTX1080 | 1 | FP16 | 608x608 | 38.5 |
 | YOLOv4(CSPDarknet53) | Xeon E5-2620/GTX1080 | 1 | FP16 | 608x608 | 35.7 |
 | YOLOv4(CSPDarknet53) | Xeon E5-2620/GTX1080 | 4 | FP16 | 608x608 | 40.9 |
 | YOLOv4(CSPDarknet53) | Xeon E5-2620/GTX1080 | 8 | FP16 | 608x608 | 41.3 | 
diff --git a/arcface/README.md b/arcface/README.md
index c16ae75..c0df3ec 100644
--- a/arcface/README.md
+++ b/arcface/README.md
@@ -4,7 +4,7 @@ The mxnet implementation is from [deepinsight/insightface.](https://github.com/d
 
 The pretrained model is [LResNet50E-IR,ArcFace@ms1m-refine-v1.](https://github.com/deepinsight/insightface/wiki/Model-Zoo#32-lresnet50e-irarcfacems1m-refine-v1)
 
-The two images used in this project are joey0.ppm and joey1.ppm, download them from [Google Drive.](https://drive.google.com/drive/folders/1ctqpkRCRKyBZRCNwo9Uq4eUoMRLtFq1e)
+The two input images used in this project are joey0.ppm and joey1.ppm, download them from [Google Drive.](https://drive.google.com/drive/folders/1ctqpkRCRKyBZRCNwo9Uq4eUoMRLtFq1e). The input image is 112x112, and generated from `get_input()` in `insightface/deploy/face_model.py`, which is cropped and aligned face image.
 
 <p align="center">
 <img src="https://user-images.githubusercontent.com/15235574/83122953-f45f8d80-a106-11ea-84b0-4f6ff91b5924.jpg">
diff --git a/yolov3-spp/CMakeLists.txt b/yolov3-spp/CMakeLists.txt
index a466470..4774663 100644
--- a/yolov3-spp/CMakeLists.txt
+++ b/yolov3-spp/CMakeLists.txt
@@ -32,8 +32,8 @@ cuda_add_library(yololayer SHARED ${PROJECT_SOURCE_DIR}/yololayer.cu)
 find_package(OpenCV)
 include_directories(OpenCV_INCLUDE_DIRS)
 
-add_executable(yolov3-spp ${PROJECT_SOURCE_DIR}/plugin_factory.cpp ${PROJECT_SOURCE_DIR}/yolov3-spp.cpp)
-target_link_libraries(yolov3-spp nvinfer nvinfer_plugin)
+add_executable(yolov3-spp ${PROJECT_SOURCE_DIR}/yolov3-spp.cpp)
+target_link_libraries(yolov3-spp nvinfer)
 target_link_libraries(yolov3-spp cudart)
 target_link_libraries(yolov3-spp yololayer)
 target_link_libraries(yolov3-spp ${OpenCV_LIBS})
diff --git a/yolov3-spp/common.h b/yolov3-spp/common.h
deleted file mode 100644
index 3b9c30c..0000000
--- a/yolov3-spp/common.h
+++ /dev/null
@@ -1,356 +0,0 @@
-#ifndef _TRT_COMMON_H_
-#define _TRT_COMMON_H_
-#include "NvInfer.h"
-#include "NvOnnxConfig.h"
-#include "NvOnnxParser.h"
-#include <cuda_runtime_api.h>
-#include <algorithm>
-#include <cassert>
-#include <fstream>
-#include <iostream>
-#include <iterator>
-#include <map>
-#include <memory>
-#include <numeric>
-#include <string>
-#include <vector>
-#include <cstring>
-#include <cmath>
-
-using namespace std;
-
-#define CHECK(status)                             \
-    do                                            \
-    {                                             \
-        auto ret = (status);                      \
-        if (ret != 0)                             \
-        {                                         \
-            std::cout << "Cuda failure: " << ret; \
-            abort();                              \
-        }                                         \
-    } while (0)
-
-constexpr long double operator"" _GB(long double val) { return val * (1 << 30); }
-constexpr long double operator"" _MB(long double val) { return val * (1 << 20); }
-constexpr long double operator"" _KB(long double val) { return val * (1 << 10); }
-
-// These is necessary if we want to be able to write 1_GB instead of 1.0_GB.
-// Since the return type is signed, -1_GB will work as expected.
-constexpr long long int operator"" _GB(long long unsigned int val) { return val * (1 << 30); }
-constexpr long long int operator"" _MB(long long unsigned int val) { return val * (1 << 20); }
-constexpr long long int operator"" _KB(long long unsigned int val) { return val * (1 << 10); }
-
-// 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};
-};
-
-// Locate path to file, given its filename or filepath suffix and possible dirs it might lie in
-// Function will also walk back MAX_DEPTH dirs from CWD to check for such a file path
-inline std::string locateFile(const std::string& filepathSuffix, const std::vector<std::string>& directories)
-{
-    const int MAX_DEPTH{10};
-    bool found{false};
-    std::string filepath;
-
-    for (auto& dir : directories)
-    {
-        filepath = dir + filepathSuffix;
-
-        for (int i = 0; i < MAX_DEPTH && !found; i++)
-        {
-            std::ifstream checkFile(filepath);
-            found = checkFile.is_open();
-            if (found) break;
-            filepath = "../" + filepath; // Try again in parent dir
-        }
-
-        if (found)
-        {
-            break;
-        }
-
-        filepath.clear();
-    }
-
-    if (filepath.empty()) {
-        std::string directoryList = std::accumulate(directories.begin() + 1, directories.end(), directories.front(),
-            [](const std::string& a, const std::string& b) { return a + "\n\t" + b; });
-        throw std::runtime_error("Could not find " + filepathSuffix + " in data directories:\n\t" + directoryList);
-    }
-    return filepath;
-}
-
-inline void readPGMFile(const std::string& fileName, uint8_t* buffer, int inH, int inW)
-{
-    std::ifstream infile(fileName, std::ifstream::binary);
-    assert(infile.is_open() && "Attempting to read from a file that is not open.");
-    std::string magic, h, w, max;
-    infile >> magic >> h >> w >> max;
-    infile.seekg(1, infile.cur);
-    infile.read(reinterpret_cast<char*>(buffer), inH * inW);
-}
-
-namespace samples_common
-{
-
-inline void* safeCudaMalloc(size_t memSize)
-{
-    void* deviceMem;
-    CHECK(cudaMalloc(&deviceMem, memSize));
-    if (deviceMem == nullptr)
-    {
-        std::cerr << "Out of memory" << std::endl;
-        exit(1);
-    }
-    return deviceMem;
-}
-
-inline bool isDebug()
-{
-    return (std::getenv("TENSORRT_DEBUG") ? true : false);
-}
-
-struct InferDeleter
-{
-    template <typename T>
-    void operator()(T* obj) const
-    {
-        if (obj) {
-            obj->destroy();
-        }
-    }
-};
-
-template <typename T>
-inline std::shared_ptr<T> infer_object(T* obj)
-{
-    if (!obj) {
-        throw std::runtime_error("Failed to create object");
-    }
-    return std::shared_ptr<T>(obj, InferDeleter());
-}
-
-template <class Iter>
-inline std::vector<size_t> argsort(Iter begin, Iter end, bool reverse = false)
-{
-    std::vector<size_t> inds(end - begin);
-    std::iota(inds.begin(), inds.end(), 0);
-    if (reverse) {
-        std::sort(inds.begin(), inds.end(), [&begin](size_t i1, size_t i2) {
-            return begin[i2] < begin[i1];
-        });
-    }
-    else
-    {
-        std::sort(inds.begin(), inds.end(), [&begin](size_t i1, size_t i2) {
-            return begin[i1] < begin[i2];
-        });
-    }
-    return inds;
-}
-
-inline bool readReferenceFile(const std::string& fileName, std::vector<std::string>& refVector)
-{
-    std::ifstream infile(fileName);
-    if (!infile.is_open()) {
-        cout << "ERROR: readReferenceFile: Attempting to read from a file that is not open." << endl;
-        return false;
-    }
-    std::string line;
-    while (std::getline(infile, line)) {
-        if (line.empty()) continue;
-        refVector.push_back(line);
-    }
-    infile.close();
-    return true;
-}
-
-template <typename result_vector_t>
-inline std::vector<std::string> classify(const vector<string>& refVector, const result_vector_t& output, const size_t topK)
-{
-    auto inds = samples_common::argsort(output.cbegin(), output.cend(), true);
-    std::vector<std::string> result;
-    for (size_t k = 0; k < topK; ++k) {
-        result.push_back(refVector[inds[k]]);
-    }
-    return result;
-}
-
-//...LG returns top K indices, not values.
-template <typename T>
-inline vector<size_t> topK(const vector<T> inp, const size_t k)
-{
-    vector<size_t> result;
-    std::vector<size_t> inds = samples_common::argsort(inp.cbegin(), inp.cend(), true);
-    result.assign(inds.begin(), inds.begin()+k);
-    return result;
-}
-
-template <typename T>
-inline bool readASCIIFile(const string& fileName, const size_t size, vector<T>& out)
-{
-    std::ifstream infile(fileName);
-    if (!infile.is_open()) {
-        cout << "ERROR readASCIIFile: Attempting to read from a file that is not open." << endl;
-        return false;
-    }
-    out.clear();
-    out.reserve(size);
-    out.assign(std::istream_iterator<T>(infile), std::istream_iterator<T>());
-    infile.close();
-    return true;
-}
-
-template <typename T>
-inline bool writeASCIIFile(const string& fileName, const vector<T>& in)
-{
-    std::ofstream outfile(fileName);
-    if (!outfile.is_open()) {
-        cout << "ERROR: writeASCIIFile: Attempting to write to a file that is not open." << endl;
-        return false;
-    }
-    for (auto fn : in) {
-        outfile << fn << " ";
-    }
-    outfile.close();
-    return true;
-}
-
-inline void print_version()
-{
-//... This can be only done after statically linking this support into parserONNX.library
-#if 0
-    std::cout << "Parser built against:" << std::endl;
-    std::cout << "  ONNX IR version:  " << nvonnxparser::onnx_ir_version_string(onnx::IR_VERSION) << std::endl;
-#endif
-    std::cout << "  TensorRT version: "
-              << NV_TENSORRT_MAJOR << "."
-              << NV_TENSORRT_MINOR << "."
-              << NV_TENSORRT_PATCH << "."
-              << NV_TENSORRT_BUILD << std::endl;
-}
-
-inline string getFileType(const string& filepath)
-{
-    return filepath.substr(filepath.find_last_of(".") + 1);
-}
-
-inline string toLower(const string& inp)
-{
-    string out = inp;
-    std::transform(out.begin(), out.end(), out.begin(), ::tolower);
-    return out;
-}
-
-inline unsigned int getElementSize(nvinfer1::DataType t)
-{
-    switch (t)
-    {
-        case nvinfer1::DataType::kINT32: return 4;
-        case nvinfer1::DataType::kFLOAT: return 4;
-        case nvinfer1::DataType::kHALF: return 2;
-        case nvinfer1::DataType::kINT8: return 1;
-    }
-    throw std::runtime_error("Invalid DataType.");
-    return 0;
-}
-
-inline int64_t volume(const nvinfer1::Dims& d)
-{
-    return std::accumulate(d.d, d.d + d.nbDims, 1, std::multiplies<int64_t>());
-}
-
-// Struct to maintain command-line arguments.
-struct Args
-{
-    bool runInInt8 = false;
-};
-
-// Populates the Args struct with the provided command-line parameters.
-inline void parseArgs(Args& args, int argc, char* argv[])
-{
-    if (argc >= 1)
-    {
-        for (int i = 1; i < argc; ++i)
-        {
-            if (!strcmp(argv[i], "--int8")) args.runInInt8 = true;
-        }
-    }
-}
-
-template <int C, int H, int W>
-struct PPM
-{
-    std::string magic, fileName;
-    int h, w, max;
-    uint8_t buffer[C * H * W];
-};
-
-struct BBox
-{
-    float x1, y1, x2, y2;
-};
-
-template <int C, int H, int W>
-inline void writePPMFileWithBBox(const std::string& filename, PPM<C, H, W>& ppm, const BBox& bbox)
-{
-    std::ofstream outfile("./" + filename, std::ofstream::binary);
-    assert(!outfile.fail());
-    outfile << "P6" << "\n" << ppm.w << " " << ppm.h << "\n" << ppm.max << "\n";
-    auto round = [](float x) -> int { return int(std::floor(x + 0.5f)); };
-    const int x1 = std::min(std::max(0, round(int(bbox.x1))), W - 1);
-    const int x2 = std::min(std::max(0, round(int(bbox.x2))), W - 1);
-    const int y1 = std::min(std::max(0, round(int(bbox.y1))), H - 1);
-    const int y2 = std::min(std::max(0, round(int(bbox.y2))), H - 1);
-    for (int x = x1; x <= x2; ++x)
-    {
-        // bbox top border
-        ppm.buffer[(y1 * ppm.w + x) * 3] = 255;
-        ppm.buffer[(y1 * ppm.w + x) * 3 + 1] = 0;
-        ppm.buffer[(y1 * ppm.w + x) * 3 + 2] = 0;
-        // bbox bottom border
-        ppm.buffer[(y2 * ppm.w + x) * 3] = 255;
-        ppm.buffer[(y2 * ppm.w + x) * 3 + 1] = 0;
-        ppm.buffer[(y2 * ppm.w + x) * 3 + 2] = 0;
-    }
-    for (int y = y1; y <= y2; ++y)
-    {
-        // bbox left border
-        ppm.buffer[(y * ppm.w + x1) * 3] = 255;
-        ppm.buffer[(y * ppm.w + x1) * 3 + 1] = 0;
-        ppm.buffer[(y * ppm.w + x1) * 3 + 2] = 0;
-        // bbox right border
-        ppm.buffer[(y * ppm.w + x2) * 3] = 255;
-        ppm.buffer[(y * ppm.w + x2) * 3 + 1] = 0;
-        ppm.buffer[(y * ppm.w + x2) * 3 + 2] = 0;
-    }
-    outfile.write(reinterpret_cast<char*>(ppm.buffer), ppm.w * ppm.h * 3);
-}
-
-} // namespace samples_common
-
-#endif // _TRT_COMMON_H_
diff --git a/yolov3-spp/logging.h b/yolov3-spp/logging.h
new file mode 100644
index 0000000..602b69f
--- /dev/null
+++ b/yolov3-spp/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/yolov3-spp/plugin_factory.cpp b/yolov3-spp/plugin_factory.cpp
deleted file mode 100644
index e8ceb4d..0000000
--- a/yolov3-spp/plugin_factory.cpp
+++ /dev/null
@@ -1,17 +0,0 @@
-#include "common.h"
-#include "plugin_factory.h"
-#include "NvInferPlugin.h"
-#include "yololayer.h"
-
-using namespace nvinfer1;
-using nvinfer1::PluginFactory;
-
-IPlugin* PluginFactory::createPlugin(const char* layerName, const void* serialData, size_t serialLength) {
-    IPlugin *plugin = nullptr;
-    if (strstr(layerName, "leaky") != NULL) {
-        plugin = plugin::createPReLUPlugin(serialData, serialLength);
-    } else if (strstr(layerName, "yolo") != NULL) {
-        plugin = new YoloLayerPlugin(serialData, serialLength);
-    }
-    return plugin;
-}
diff --git a/yolov3-spp/plugin_factory.h b/yolov3-spp/plugin_factory.h
deleted file mode 100644
index 0be0225..0000000
--- a/yolov3-spp/plugin_factory.h
+++ /dev/null
@@ -1,12 +0,0 @@
-#ifndef MY_PLUGIN_FACTORY_H
-#define MY_PLUGIN_FACTORY_H
-#include <NvInfer.h>
-
-namespace nvinfer1 {
-class PluginFactory : public IPluginFactory {
-    public:
-        IPlugin* createPlugin(const char* layerName, const void* serialData, size_t serialLength) override;
-};
-
-}
-#endif
diff --git a/yolov3-spp/yololayer.cu b/yolov3-spp/yololayer.cu
index 3804b45..722cb27 100644
--- a/yolov3-spp/yololayer.cu
+++ b/yolov3-spp/yololayer.cu
@@ -4,7 +4,7 @@ using namespace Yolo;
 
 namespace nvinfer1
 {
-    YoloLayerPlugin::YoloLayerPlugin(const int cudaThread /*= 512*/):mThreadCount(cudaThread)
+    YoloLayerPlugin::YoloLayerPlugin()
     {
         mClassCount = CLASS_NUM;
         mYoloKernel.clear();
@@ -18,7 +18,7 @@ namespace nvinfer1
     YoloLayerPlugin::~YoloLayerPlugin()
     {
     }
-    
+
     // create the plugin at runtime from a byte stream
     YoloLayerPlugin::YoloLayerPlugin(const void* data, size_t length)
     {
@@ -35,7 +35,7 @@ namespace nvinfer1
         assert(d == a + length);
     }
 
-    void YoloLayerPlugin::serialize(void* buffer)
+    void YoloLayerPlugin::serialize(void* buffer) const
     {
         using namespace Tn;
         char* d = static_cast<char*>(buffer), *a = d;
@@ -49,34 +49,89 @@ namespace nvinfer1
         assert(d == a + getSerializationSize());
     }
     
-    size_t YoloLayerPlugin::getSerializationSize()
+    size_t YoloLayerPlugin::getSerializationSize() const
     {  
         return sizeof(mClassCount) + sizeof(mThreadCount) + sizeof(mKernelCount)  + sizeof(Yolo::YoloKernel) * mYoloKernel.size();
     }
 
     int YoloLayerPlugin::initialize()
     { 
-        int totalCount = 0;
-        for(const auto& yolo : mYoloKernel)
-            totalCount += (LOCATIONS + 1) * yolo.width*yolo.height * CHECK_COUNT;
-
-        totalCount = 0;//detection count
-        for(const auto& yolo : mYoloKernel)
-            totalCount += yolo.width*yolo.height * CHECK_COUNT;
         return 0;
     }
     
     Dims YoloLayerPlugin::getOutputDimensions(int index, const Dims* inputs, int nbInputDims)
     {
         //output the result to channel
-        int totalCount = 0;
-        for(const auto& yolo : mYoloKernel)
-            totalCount += yolo.width*yolo.height * CHECK_COUNT * sizeof(Detection) / sizeof(float);
+        int totalsize = MAX_OUTPUT_BBOX_COUNT * sizeof(Detection) / sizeof(float);
 
-        return Dims3(totalCount + 1, 1, 1);
+        return Dims3(totalsize + 1, 1, 1);
     }
 
-    __device__ float Logist(float data){ return 1./(1. + exp(-data)); };
+    // 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) {
@@ -85,26 +140,27 @@ namespace nvinfer1
         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;
-        //int info_len_o = 7;
-        int input_col = idx;
-        //int out_row = input_col;
+        const float* curInput = input + bnIdx * (info_len_i * total_grid * CHECK_COUNT);
 
         for (int k = 0; k < 3; ++k) {
             int class_id = 0;
             float max_cls_prob = 0.0;
             for (int i = 5; i < info_len_i; ++i) {
-                float p = Logist(input[input_col + k * info_len_i * total_grid + i * total_grid]);
+                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 box_prob = Logist(input[input_col + k * info_len_i * total_grid + 4 * total_grid]);
+            float box_prob = Logist(curInput[idx + k * info_len_i * total_grid + 4 * total_grid]);
             if (max_cls_prob < IGNORE_THRESH || box_prob < IGNORE_THRESH) continue;
 
-            float *res_count = output;
+            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);
 
@@ -112,37 +168,32 @@ namespace nvinfer1
             int col = idx % yoloWidth;
 
             //Location
-            det->bbox[0] = (col + Logist(input[input_col + k * info_len_i * total_grid + 0 * total_grid])) * INPUT_W / yoloWidth;
-            det->bbox[1] = (row + Logist(input[input_col + k * info_len_i * total_grid + 1 * total_grid])) * INPUT_H / yoloHeight;
-            det->bbox[2] = exp(input[input_col + k * info_len_i * total_grid + 2 * total_grid]) * anchors[2*k];
-            det->bbox[3] = exp(input[input_col + k * info_len_i * total_grid + 3 * total_grid]) * anchors[2*k + 1];
+            det->bbox[0] = (col + Logist(curInput[idx + k * info_len_i * total_grid + 0 * total_grid])) * INPUT_W / yoloWidth;
+            det->bbox[1] = (row + Logist(curInput[idx + k * info_len_i * total_grid + 1 * total_grid])) * INPUT_H / yoloHeight;
+            det->bbox[2] = expf(curInput[idx + k * info_len_i * total_grid + 2 * total_grid]) * anchors[2*k];
+            det->bbox[3] = expf(curInput[idx + k * info_len_i * total_grid + 3 * total_grid]) * anchors[2*k + 1];
             det->det_confidence = box_prob;
             det->class_id = class_id;
             det->class_confidence = max_cls_prob;
         }
     }
-   
-    void YoloLayerPlugin::forwardGpu(const float *const * inputs,float * output,cudaStream_t stream,int batchSize) {
+
+    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;
-        for (unsigned int i = 0;i< mYoloKernel.size();++i)
-        {
-            const auto& yolo = mYoloKernel[i];
-            outputElem += yolo.width*yolo.height * CHECK_COUNT * sizeof(Detection) / sizeof(float);
-        }
+        int outputElem = 1 + MAX_OUTPUT_BBOX_COUNT * sizeof(Detection) / sizeof(float);
 
-        for(int idx = 0 ;idx < batchSize;++idx)
+        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 < 256)
+            if (numElem < mThreadCount)
                 mThreadCount = numElem;
             CUDA_CHECK(cudaMemcpy(devAnchor, yolo.anchors, AnchorLen, cudaMemcpyHostToDevice));
             CalDetection<<< (yolo.width*yolo.height*batchSize + mThreadCount - 1) / mThreadCount, mThreadCount>>>
@@ -158,9 +209,51 @@ namespace nvinfer1
         //assert(batchSize == 1);
         //GPU
         //CUDA_CHECK(cudaStreamSynchronize(stream));
-        forwardGpu((const float *const *)inputs,(float *)outputs[0],stream,batchSize);
+        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/yolov3-spp/yololayer.h b/yolov3-spp/yololayer.h
index 10ec336..37ad03e 100644
--- a/yolov3-spp/yololayer.h
+++ b/yolov3-spp/yololayer.h
@@ -4,7 +4,6 @@
 #include <assert.h>
 #include <cmath>
 #include <string.h>
-#include <cudnn.h>
 #include <cublas_v2.h>
 #include "NvInfer.h"
 #include "Utils.h"
@@ -14,9 +13,10 @@ 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 = 256;
-    static constexpr int INPUT_W = 416;
+    static constexpr int INPUT_H = 608;
+    static constexpr int INPUT_W = 608;
 
     struct YoloKernel
     {
@@ -25,17 +25,17 @@ namespace Yolo
         float anchors[CHECK_COUNT*2];
     };
 
-    static YoloKernel yolo1 = {
+    static constexpr YoloKernel yolo1 = {
         INPUT_W / 32,
         INPUT_H / 32,
         {116,90,  156,198,  373,326}
     };
-    static YoloKernel yolo2 = {
+    static constexpr YoloKernel yolo2 = {
         INPUT_W / 16,
         INPUT_H / 16,
         {30,61,  62,45,  59,119}
     };
-    static YoloKernel yolo3 = {
+    static constexpr YoloKernel yolo3 = {
         INPUT_W / 8,
         INPUT_H / 8,
         {10,13,  16,30,  33,23}
@@ -54,48 +54,106 @@ namespace Yolo
 
 namespace nvinfer1
 {
-    class YoloLayerPlugin: public IPluginExt
+    class YoloLayerPlugin: public IPluginV2IOExt
     {
-    public:
-        explicit YoloLayerPlugin(const int cudaThread = 256);
-        YoloLayerPlugin(const void* data, size_t length);
+        public:
+            explicit YoloLayerPlugin();
+            YoloLayerPlugin(const void* data, size_t length);
 
-        ~YoloLayerPlugin();
+            ~YoloLayerPlugin();
 
-        int getNbOutputs() const override
-        {
-            return 1;
-        }
+            int getNbOutputs() const override
+            {
+                return 1;
+            }
 
-        Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override;
+            Dims getOutputDimensions(int index, const Dims* inputs, int nbInputDims) override;
 
-        bool supportsFormat(DataType type, PluginFormat format) const override { 
-            return type == DataType::kFLOAT && format == PluginFormat::kNCHW; 
-        }
+            int initialize() override;
 
-        void configureWithFormat(const Dims* inputDims, int nbInputs, const Dims* outputDims, int nbOutputs, DataType type, PluginFormat format, int maxBatchSize) override {};
+            virtual void terminate() override {};
 
-        int initialize() override;
+            virtual size_t getWorkspaceSize(int maxBatchSize) const override { return 0;}
 
-        virtual void terminate() override {};
+            virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override;
 
-        virtual size_t getWorkspaceSize(int maxBatchSize) const override { return 0;}
+            virtual size_t getSerializationSize() const override;
 
-        virtual int enqueue(int batchSize, const void*const * inputs, void** outputs, void* workspace, cudaStream_t stream) override;
+            virtual void serialize(void* buffer) const override;
 
-        virtual size_t getSerializationSize() 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;
+            }
 
-        virtual void serialize(void* buffer) override;
+            const char* getPluginType() const override;
 
-        void forwardGpu(const float *const * inputs,float * output, cudaStream_t stream,int batchSize = 1);
+            const char* getPluginVersion() const override;
 
-    private:
-        int mClassCount;
-        int mKernelCount;
-        std::vector<Yolo::YoloKernel> mYoloKernel;
-        int mThreadCount;
-        //int mDetNum;
+            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/yolov3-spp/yolov3-spp.cpp b/yolov3-spp/yolov3-spp.cpp
index a18fb3f..1bc0611 100644
--- a/yolov3-spp/yolov3-spp.cpp
+++ b/yolov3-spp/yolov3-spp.cpp
@@ -1,19 +1,29 @@
-#include "NvInfer.h"
-#include "NvInferPlugin.h"
-#include "cuda_runtime_api.h"
-#include "common.h"
 #include <fstream>
 #include <iostream>
 #include <map>
 #include <sstream>
 #include <vector>
 #include <chrono>
-#include "plugin_factory.h"
-#include "yololayer.h"
 #include <opencv2/opencv.hpp>
 #include <dirent.h>
+#include "NvInfer.h"
+#include "cuda_runtime_api.h"
+#include "logging.h"
+#include "yololayer.h"
 
-//#define USE_FP16  // comment out this if want to use FP32
+#define CHECK(status) \
+    do\
+    {\
+        auto ret = (status);\
+        if (ret != 0)\
+        {\
+            std::cerr << "Cuda failure: " << ret << std::endl;\
+            abort();\
+        }\
+    } while (0)
+
+
+#define USE_FP16  // comment out this if want to use FP32
 #define DEVICE 0  // GPU id
 #define NMS_THRESH 0.4
 #define BBOX_CONF_THRESH 0.5
@@ -27,6 +37,7 @@ static const int OUTPUT_SIZE = 1000 * 7 + 1;  // we assume the yololayer outputs
 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;
@@ -78,10 +89,10 @@ cv::Rect get_rect(cv::Mat& img, float bbox[4]) {
 
 float iou(float lbox[4], float rbox[4]) {
     float interBox[] = {
-        max(lbox[0] - lbox[2]/2.f , rbox[0] - rbox[2]/2.f), //left
-        min(lbox[0] + lbox[2]/2.f , rbox[0] + rbox[2]/2.f), //right
-        max(lbox[1] - lbox[3]/2.f , rbox[1] - rbox[3]/2.f), //top
-        min(lbox[1] + lbox[3]/2.f , rbox[1] + rbox[3]/2.f), //bottom
+        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])
@@ -167,7 +178,6 @@ IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, W
     float *mean = (float*)weightMap[lname + ".running_mean"].values;
     float *var = (float*)weightMap[lname + ".running_var"].values;
     int len = weightMap[lname + ".running_var"].count;
-    std::cout << "len " << len << std::endl;
 
     float *scval = reinterpret_cast<float*>(malloc(sizeof(float) * len));
     for (int i = 0; i < len; i++) {
@@ -196,29 +206,25 @@ IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map<std::string, W
 }
 
 ILayer* convBnLeaky(INetworkDefinition *network, std::map<std::string, Weights>& weightMap, ITensor& input,  int outch, int ksize, int s, int p, int linx) {
-    std::cout << linx << std::endl;
     Weights emptywts{DataType::kFLOAT, nullptr, 0};
-    IConvolutionLayer* conv1 = network->addConvolution(input, outch, DimsHW{ksize, ksize}, weightMap["module_list." + std::to_string(linx) + ".Conv2d.weight"], emptywts);
+    IConvolutionLayer* conv1 = network->addConvolutionNd(input, outch, DimsHW{ksize, ksize}, weightMap["module_list." + std::to_string(linx) + ".Conv2d.weight"], emptywts);
     assert(conv1);
-    conv1->setStride(DimsHW{s, s});
-    conv1->setPadding(DimsHW{p, p});
+    conv1->setStrideNd(DimsHW{s, s});
+    conv1->setPaddingNd(DimsHW{p, p});
 
     IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *conv1->getOutput(0), "module_list." + std::to_string(linx) + ".BatchNorm2d", 1e-5);
 
-    ITensor* inputTensors[] = {bn1->getOutput(0)};
-    //LeakyPlugin *lr = new LeakyPlugin();
-    auto lr = plugin::createPReLUPlugin(0.1);
-    auto lr1 = network->addPlugin(inputTensors, 1, *lr);
-    assert(lr1);
-    lr1->setName(("leaky" + std::to_string(linx)).c_str());
-    return lr1;
+    auto lr = network->addActivation(*bn1->getOutput(0), ActivationType::kLEAKY_RELU);
+    lr->setAlpha(0.1);
+
+    return lr;
 }
 
 // Creat the engine using only the API and not any parser.
-ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType dt) {
-    INetworkDefinition* network = builder->createNetwork();
+ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig* config, DataType dt) {
+    INetworkDefinition* network = builder->createNetworkV2(0U);
 
-    // Create input tensor of shape { 1, 1, 32, 32 } with name INPUT_BLOB_NAME
+    // Create input tensor of shape {3, INPUT_H, INPUT_W} with name INPUT_BLOB_NAME
     ITensor* data = network->addInput(INPUT_BLOB_NAME, dt, Dims3{3, INPUT_H, INPUT_W});
     assert(data);
 
@@ -305,15 +311,15 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
     auto lr76 = convBnLeaky(network, weightMap, *lr75->getOutput(0), 1024, 3, 1, 1, 76);
     auto lr77 = convBnLeaky(network, weightMap, *lr76->getOutput(0), 512, 1, 1, 0, 77);
     
-    auto pool78 = network->addPooling(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{5,5});
-    pool78->setPadding(DimsHW{2, 2});
-    pool78->setStride(DimsHW{1, 1});
-    auto pool80 = network->addPooling(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{9,9});
-    pool80->setPadding(DimsHW{4, 4});
-    pool80->setStride(DimsHW{1, 1});
-    auto pool82 = network->addPooling(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{13,13});
-    pool82->setPadding(DimsHW{6, 6});
-    pool82->setStride(DimsHW{1, 1});
+    auto pool78 = network->addPoolingNd(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{5,5});
+    pool78->setPaddingNd(DimsHW{2, 2});
+    pool78->setStrideNd(DimsHW{1, 1});
+    auto pool80 = network->addPoolingNd(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{9,9});
+    pool80->setPaddingNd(DimsHW{4, 4});
+    pool80->setStrideNd(DimsHW{1, 1});
+    auto pool82 = network->addPoolingNd(*lr77->getOutput(0), PoolingType::kMAX, DimsHW{13,13});
+    pool82->setPaddingNd(DimsHW{6, 6});
+    pool82->setStrideNd(DimsHW{1, 1});
 
     ITensor* inputTensors83[] = {pool82->getOutput(0), pool80->getOutput(0), pool78->getOutput(0), lr77->getOutput(0)};
     auto cat83 = network->addConcatenation(inputTensors83, 4);
@@ -322,7 +328,7 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
     auto lr85 = convBnLeaky(network, weightMap, *lr84->getOutput(0), 1024, 3, 1, 1, 85);
     auto lr86 = convBnLeaky(network, weightMap, *lr85->getOutput(0), 512, 1, 1, 0, 86);
     auto lr87 = convBnLeaky(network, weightMap, *lr86->getOutput(0), 1024, 3, 1, 1, 87);
-    IConvolutionLayer* conv88 = network->addConvolution(*lr87->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.88.Conv2d.weight"], weightMap["module_list.88.Conv2d.bias"]);
+    IConvolutionLayer* conv88 = network->addConvolutionNd(*lr87->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.88.Conv2d.weight"], weightMap["module_list.88.Conv2d.bias"]);
     assert(conv88);
     auto lr91 = convBnLeaky(network, weightMap, *lr86->getOutput(0), 256, 1, 1, 0, 91);
 
@@ -331,9 +337,9 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
         deval[i] = 1.0;
     }
     Weights deconvwts92{DataType::kFLOAT, deval, 256 * 2 * 2};
-    IDeconvolutionLayer* deconv92 = network->addDeconvolution(*lr91->getOutput(0), 256, DimsHW{2, 2}, deconvwts92, emptywts);
+    IDeconvolutionLayer* deconv92 = network->addDeconvolutionNd(*lr91->getOutput(0), 256, DimsHW{2, 2}, deconvwts92, emptywts);
     assert(deconv92);
-    deconv92->setStride(DimsHW{2, 2});
+    deconv92->setStrideNd(DimsHW{2, 2});
     deconv92->setNbGroups(256);
     weightMap["deconv92"] = deconvwts92;
     
@@ -345,13 +351,13 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
     auto lr97 = convBnLeaky(network, weightMap, *lr96->getOutput(0), 512, 3, 1, 1, 97);
     auto lr98 = convBnLeaky(network, weightMap, *lr97->getOutput(0), 256, 1, 1, 0, 98);
     auto lr99 = convBnLeaky(network, weightMap, *lr98->getOutput(0), 512, 3, 1, 1, 99);
-    IConvolutionLayer* conv100 = network->addConvolution(*lr99->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.100.Conv2d.weight"], weightMap["module_list.100.Conv2d.bias"]);
+    IConvolutionLayer* conv100 = network->addConvolutionNd(*lr99->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.100.Conv2d.weight"], weightMap["module_list.100.Conv2d.bias"]);
     assert(conv100);
     auto lr103 = convBnLeaky(network, weightMap, *lr98->getOutput(0), 128, 1, 1, 0, 103);
     Weights deconvwts104{DataType::kFLOAT, deval, 128 * 2 * 2};
-    IDeconvolutionLayer* deconv104 = network->addDeconvolution(*lr103->getOutput(0), 128, DimsHW{2, 2}, deconvwts104, emptywts);
+    IDeconvolutionLayer* deconv104 = network->addDeconvolutionNd(*lr103->getOutput(0), 128, DimsHW{2, 2}, deconvwts104, emptywts);
     assert(deconv104);
-    deconv104->setStride(DimsHW{2, 2});
+    deconv104->setStrideNd(DimsHW{2, 2});
     deconv104->setNbGroups(128);
     ITensor* inputTensors1[] = {deconv104->getOutput(0), ew36->getOutput(0)};
     auto cat105 = network->addConcatenation(inputTensors1, 2);
@@ -361,26 +367,27 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
     auto lr109 = convBnLeaky(network, weightMap, *lr108->getOutput(0), 256, 3, 1, 1, 109);
     auto lr110 = convBnLeaky(network, weightMap, *lr109->getOutput(0), 128, 1, 1, 0, 110);
     auto lr111 = convBnLeaky(network, weightMap, *lr110->getOutput(0), 256, 3, 1, 1, 111);
-    IConvolutionLayer* conv112 = network->addConvolution(*lr111->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.112.Conv2d.weight"], weightMap["module_list.112.Conv2d.bias"]);
+    IConvolutionLayer* conv112 = network->addConvolutionNd(*lr111->getOutput(0), 3 * (Yolo::CLASS_NUM + 5), DimsHW{1, 1}, weightMap["module_list.112.Conv2d.weight"], weightMap["module_list.112.Conv2d.bias"]);
     assert(conv112);
-    auto yolo = new YoloLayerPlugin();
-    ITensor* inputTensors_yolo[] = {conv88->getOutput(0), conv100->getOutput(0), conv112->getOutput(0)};
-    auto yolo113 = network->addPlugin(inputTensors_yolo, 3, *yolo);
-    assert(yolo113);
-    yolo113->setName("yolo113");
 
-    yolo113->getOutput(0)->setName(OUTPUT_BLOB_NAME);
-    std::cout << "set name out" << std::endl;
-    network->markOutput(*yolo113->getOutput(0));
+    auto creator = getPluginRegistry()->getPluginCreator("YoloLayer_TRT", "1");
+    const PluginFieldCollection* pluginData = creator->getFieldNames();
+    IPluginV2 *pluginObj = creator->createPlugin("yololayer", pluginData);
+    ITensor* inputTensors_yolo[] = {conv88->getOutput(0), conv100->getOutput(0), conv112->getOutput(0)};
+    auto yolo = network->addPluginV2(inputTensors_yolo, 3, *pluginObj);
+
+    yolo->getOutput(0)->setName(OUTPUT_BLOB_NAME);
+    network->markOutput(*yolo->getOutput(0));
 
     // Build engine
     builder->setMaxBatchSize(maxBatchSize);
-    builder->setMaxWorkspaceSize(1 << 20);
+    config->setMaxWorkspaceSize(16 * (1 << 20));  // 16MB
 #ifdef USE_FP16
-    builder->setFp16Mode(true);
+    config->setFlag(BuilderFlag::kFP16);
 #endif
-    ICudaEngine* engine = builder->buildCudaEngine(*network);
-    std::cout << "build out" << std::endl;
+    std::cout << "Building engine, please wait for a while..." << std::endl;
+    ICudaEngine* engine = builder->buildEngineWithConfig(*network, *config);
+    std::cout << "Build engine successfully!" << std::endl;
 
     // Don't need the network any more
     network->destroy();
@@ -397,9 +404,10 @@ ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, DataType
 void APIToModel(unsigned int maxBatchSize, IHostMemory** modelStream) {
     // Create builder
     IBuilder* builder = createInferBuilder(gLogger);
+    IBuilderConfig* config = builder->createBuilderConfig();
 
     // Create model to populate the network, then set the outputs and create an engine
-    ICudaEngine* engine = createEngine(maxBatchSize, builder, DataType::kFLOAT);
+    ICudaEngine* engine = createEngine(maxBatchSize, builder, config, DataType::kFLOAT);
     assert(engine != nullptr);
 
     // Serialize the engine
@@ -475,7 +483,7 @@ int main(int argc, char** argv) {
         IHostMemory* modelStream{nullptr};
         APIToModel(1, &modelStream);
         assert(modelStream != nullptr);
-        std::ofstream p("yolov3-spp.engine");
+        std::ofstream p("yolov3-spp.engine", std::ios::binary);
         if (!p) {
             std::cerr << "could not open plan output file" << std::endl;
             return -1;
@@ -512,13 +520,13 @@ int main(int argc, char** argv) {
     //for (int i = 0; i < 3 * INPUT_H * INPUT_W; i++)
     //    data[i] = 1.0;
     static float prob[OUTPUT_SIZE];
-    PluginFactory pf;
     IRuntime* runtime = createInferRuntime(gLogger);
     assert(runtime != nullptr);
-    ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size, &pf);
+    ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size);
     assert(engine != nullptr);
     IExecutionContext* context = engine->createExecutionContext();
     assert(context != nullptr);
+    delete[] trtModelStream;
 
     int fcount = 0;
     for (auto f: file_names) {
@@ -536,10 +544,10 @@ int main(int argc, char** argv) {
         // Run inference
         auto start = std::chrono::system_clock::now();
         doInference(*context, data, prob, 1);
-        std::vector<Yolo::Detection> res;
-        nms(res, prob);
         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);
         for (int i=0; i<20; i++) {
             std::cout << prob[i] << ",";
         }