YOLOv5+TensorRT+Win11(Python版)_python yolo5 +tensorrt

快速上手YOLOv5

一、YOLOv5算法

1. 算法对比

（1）传统目标检测方法

（2）基于深度学习的目标检测算法

（2-1）Two-Stage（R-CNN/Fast R-CNN/Faster R-CNN）

Two-Stage：先产生候选区域，后对目标分类与位置精修，准确率高但速度较慢

（2-2）One-Stage（SSD/YOLO）

One-Stage：直接回归目标类别与位置，速度快但准确率较低

（2-3）One-Stage VS Two-Stage

YOLO v5s：模型小、推理时间短、准确率较高

---

二、环境配置

1. NVIDIA独立显卡

打开设备管理器查看是否拥有NVIDIA独立显卡：

查看NVIDIA独立显卡是否支持CUDA：

https://developer.nvidia.com/cuda-gpus

---

2. 环境要求

---

3. 安装Visual Studio

安装Visual Studio 2019勾选C++桌面开发工具：

---

4. 安装Anaconda

添加系统环境变量：

测试：

换国内源：

新建.condarc文件保存以下代码，并替换C:\Users\XXX下的.condarc文件

channels:
  - defaults
show_channel_urls: true
default_channels:
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/r
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/msys2
custom_channels:
  conda-forge: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  msys2: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  bioconda: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  menpo: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  pytorch: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  pytorch-lts: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
  simpleitk: http://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

---

5. 安装CUDA Toolkit

选择自定义

全部安装

添加系统环境变量：

上述两项是安装完成Cuda后，已自动生成的环境变量的配置，我们需要自行添加的下述变量：

CUDA_BIN_PATH: %CUDA_PATH%\bin

CUDA_LIB_PATH: %CUDA_PATH%\lib\x64

CUDA_SDK_PATH: XXX\NVIDIA Corporation\CUDA Samples\v11.1

CUDA_SDK_BIN_PATH: %CUDA_SDK_PATH%\bin\win64

CUDA_SDK_LIB_PATH: %CUDA_SDK_PATH%\common\lib\x64

并在系统变量Path中，添加一下四个信息：

%CUDA_BIN_PATH%

%CUDA_LIB_PATH%

%CUDA_SDK_BIN_PATH%

%CUDA_SDK_LIB_PATH%

测试：

nvcc -V
1

---

6. 安装cuDNN

复制XXX\cudnn-11.3-windows-x64-v8.2.1.32\cuda全部内容到XXX\NVIDIA GPU Computing Toolkit\CUDA\v11.3

---

7. 安装TensorRT

添加环境变量：

复制XXX\TensorRT-8.2.5.1\lib以下所有dll文件到XXX\NVIDIA GPU Computing Toolkit\CUDA\v11.3\bin下

安装XXXTensorRT\TensorRT-8.2.5.1\python目录下tensorrt-8.2.5.1-cp39-none-win_amd64.whl：

activate pytorch
pip install tensorrt-8.2.5.1-cp39-none-win_amd64.whl -i https://pypi.tuna.tsinghua.edu.cn/simple
1
2

---

8. 安装OpenCV

添加环境变量：

---

9. 安装CMake

添加环境变量：

---

10. 安装Pytorch

https://pytorch.org/get-started/locally/

以管理员模式在命令行中运行：

conda create -n pytorch python=3.9
activate pytorch
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
1
2
3

测试：

# cuda test
import torch
print(torch.__version__)
x = torch.Tensor([1.0])
xx = x.cuda()
print(xx)
print(torch.cuda.is_available())

# cudnn test
from torch.backends import cudnn
print(cudnn.is_acceptable(xx))
1
2
3
4
5
6
7
8
9
10
11

---

11. 安装YOLOv5必要依赖包

下载YOLOv5源代码：

进入XXX\yolov5-master目录执行：

activate pytorch
pip install -r requirements.txt
1
2

测试：

activate pytorch
python detect.py
1
2

运行结果保存在XXX\yolov5-master\runs\detect\exp下

---

12. YOLOv5使用TensorRT加速

下载YOLOv5 TensorRT加速版本：

https://github.com/wang-xinyu/tensorrtx

在XXX\tensorrtx下创建include文件夹，下载文件dirent.h放置在include文件夹中，
下载地址 https://github.com/tronkko/dirent/tree/master/include/dirent.h

修改XXX\tensorrtx\yolov5目录下CMakeLists.txt内容为（按自己的安装目录修改#1 #2 #3 #4）：

cmake_minimum_required(VERSION 2.6)

project(yolov5)  
set(OpenCV_DIR "D:/Environment/OpenCV/opencv/build")    ##1
set(OpenCV_INCLUDE_DIRS "D:/Environment/OpenCV/opencv/build/include")    ##2
set(OpenCV_LIBS "D:\\Environment\\OpenCV\\opencv\\build\\x64\\vc14\\lib\\opencv_world3416.lib")    ##3
set(TRT_DIR "D:/Environment/TensorRT/TensorRT-8.2.5.1")    ##4

add_definitions(-DAPI_EXPORTS)
add_definitions(-std=c++11)
option(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_BUILD_TYPE Debug)

set(THREADS_PREFER_PTHREAD_FLAG ON)
find_package(Threads)


# setup CUDA
find_package(CUDA REQUIRED)
message(STATUS "libraries: ${CUDA_LIBRARIES}")
message(STATUS "include path: ${CUDA_INCLUDE_DIRS}")

include_directories(${CUDA_INCLUDE_DIRS})

####
enable_language(CUDA)  # add this line, then no need to setup cuda path in vs
####
include_directories(${PROJECT_SOURCE_DIR}/include)
include_directories(${TRT_DIR}\\include)
include_directories(D:\\Environment\\TensorRT\\tensorrtx\\include)    ##5


#find_package(OpenCV)
include_directories(${OpenCV_INCLUDE_DIRS})
include_directories(${OpenCV_INCLUDE_DIRS}\\opencv2) #6

# -D_MWAITXINTRIN_H_INCLUDED for solving error: identifier "__builtin_ia32_mwaitx" is undefined
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -Wall -Ofast -D_MWAITXINTRIN_H_INCLUDED")

# setup opencv
find_package(OpenCV QUIET
    NO_MODULE
    NO_DEFAULT_PATH
    NO_CMAKE_PATH
    NO_CMAKE_ENVIRONMENT_PATH
    NO_SYSTEM_ENVIRONMENT_PATH
    NO_CMAKE_PACKAGE_REGISTRY
    NO_CMAKE_BUILDS_PATH
    NO_CMAKE_SYSTEM_PATH
    NO_CMAKE_SYSTEM_PACKAGE_REGISTRY
)


message(STATUS "OpenCV library status:")
message(STATUS "version: ${OpenCV_VERSION}")
message(STATUS "libraries: ${OpenCV_LIBS}")
message(STATUS "include path: ${OpenCV_INCLUDE_DIRS}")

include_directories(${OpenCV_INCLUDE_DIRS})
link_directories(${TRT_DIR}\\lib)
link_directories(${OpenCV_DIR}\\x64\\vc14\\lib) #8

add_executable(yolov5 ${PROJECT_SOURCE_DIR}/calibrator.cpp yolov5 ${PROJECT_SOURCE_DIR}/yolov5.cpp ${PROJECT_SOURCE_DIR}/yololayer.cu ${PROJECT_SOURCE_DIR}/yololayer.h ${PROJECT_SOURCE_DIR}/preprocess.cu ${PROJECT_SOURCE_DIR}/preprocess.h)
target_link_libraries(yolov5 "nvinfer" "nvinfer_plugin") #9
target_link_libraries(yolov5 ${OpenCV_LIBS})        #10
target_link_libraries(yolov5 ${CUDA_LIBRARIES}) #11
target_link_libraries(yolov5 Threads::Threads) #12
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68

在XXX\tensorrtx\yolov5目录下新建build文件夹

运行XXX\CMake\bin\cmake-gui.exe配置：

Where is the source code: XXX\tensorrtx\yolov5

Where to build the binaries: XXX\tensorrtx\yolov5\build

依次执行Configure -> Generate -> Open Project

下载YOLOv5 TensorRT Win10加速版本：

https://github.com/Monday-Leo/Yolov5_Tensorrt_Win10

复制XXX\Yolov5_Tensorrt_Win10-master\yolov5.cpp文件内容，替换XXX\tensorrtx\yolov5\yolov5.cpp内