Centos7 安装pytorch环境并部署yolov5

1、安装 python，下次补充

2、安装配置虚拟环境

1、在根目录 新建虚拟环境目录 envs

[root@glnode02 /]# mkdir envs

2、下载 virtualanv 和 virtualenvwrapper

[root@glnode02 /]# pip3 install virtualenv
[root@glnode02 /]# pip3 install virtualenvwrapper

3、配置环境变量

[root@glnode02 /]# find / -name 'virtualenvwrapper.sh'
/usr/local/bin/virtualenvwrapper.sh

修改环境配置

[root@glnode02 /]# # vim usr/local/bin/virtualenvwrapper.sh 
......
if [ "${VIRTUALENVWRAPPER_PYTHON:-}" = "" ]
then
    # python后边加个3
    VIRTUALENVWRAPPER_PYTHON="$(command \which python3)"
fi

打开bashrc文件，末尾添加三行

[root@glnode02 /]# vim ~/.bashrc
......
export WORKON_HOME=/envs
export VIRTUELANVWRAPPER_PYTHON=/usr/bin/python3 #新安装python时添加这一行
source /usr/local/bin/virtualenvwrapper.sh

[root@glnode02 /]# source ~/.bashrc 
virtualenvwrapper.user_scripts creating /envs/premkproject
virtualenvwrapper.user_scripts creating /envs/postmkproject
virtualenvwrapper.user_scripts creating /envs/initialize
virtualenvwrapper.user_scripts creating /envs/premkvirtualenv
virtualenvwrapper.user_scripts creating /envs/postmkvirtualenv
virtualenvwrapper.user_scripts creating /envs/prermvirtualenv
virtualenvwrapper.user_scripts creating /envs/postrmvirtualenv
virtualenvwrapper.user_scripts creating /envs/predeactivate
virtualenvwrapper.user_scripts creating /envs/postdeactivate
virtualenvwrapper.user_scripts creating /envs/preactivate
virtualenvwrapper.user_scripts creating /envs/postactivate
virtualenvwrapper.user_scripts creating /envs/get_env_details
[root@glnode02 /]#

4、创建、删除虚拟环境

[root@glnode02 /]# workon # 查看虚拟环境
[root@glnode02 /]# mkvirtualenv -p python3 pytorch # 新建名为pytorch的虚拟环境 
...
[root@glnode02 /]# workon pytorch # 进入环境
(pytorch) [root@glnode02 /]# deactivate # 推出环境
[root@glnode02 /]# rmvirtualenv test_env # 删除环境

参考博客：centos7安装python3，并配置虚拟环境（超级详细教程）

3、安装torch、torchvision

1、查看自己的python版本，然后下载对应的安装包，建议使用.whl离线安装
https://download.pytorch.org/whl/torch_stable.html

(pytorch) [root@glnode02 /]# python3 -V
Python 3.6.8
You have mail in /var/spool/mail/root
(pytorch) [root@glnode02 /]# pip3 -V
pip 21.3.1 from /envs/pytorch/lib/python3.6/site-packages/pip (python 3.6)
(pytorch) [root@glnode02 /]# 

torch、torchvision版本要对应

pip install torch-1.10.1+cpu-cp36-cp36m-linux_x86_64.whl 
pip install torchvision-0.11.2+cpu-cp36-cp36m-linux_x86_64.whl

4、配置yolov5环境

pip install -r requirements.txt -i https://pypi.douban.com/simple

5、安装pytorch环境遇到的问题

1、安装opencv-python 报错

(pytorch) [root@glnode02 root]# cd yolov5/
(pytorch) [root@glnode02 yolov5]# pip install -r requirements.txt -i https://pypi.douban.com/simple
Looking in indexes: https://pypi.douban.com/simple
Collecting matplotlib>=3.2.2
  Downloading https://pypi.doubanio.com/packages/09/03/b7b30fa81cb687d1178e085d0f01111ceaea3bf81f9330c937fb6f6c8ca0/matplotlib-3.3.4-cp36-cp36m-manylinux1_x86_64.whl (11.5 MB)
     |████████████████████████████████| 11.5 MB 1.0 MB/s            
Requirement already satisfied: numpy>=1.18.5 in /envs/pytorch/lib64/python3.6/site-packages (from -r requirements.txt (line 5)) (1.19.5)
Collecting opencv-python>=4.1.1
  Downloading https://pypi.doubanio.com/packages/40/93/655af887bafece2a655998f53b9bd21ad94b0627d81d44aef35c79f40de6/opencv-python-4.7.0.72.tar.gz (91.1 MB)
     |████████████████████████████████| 91.1 MB 865 kB/s            
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... done
  ...
  ...
  Building wheels for collected packages: opencv-python
  Building wheel for opencv-python (pyproject.toml) ... error
  ERROR: Command errored out with exit status 1:
   command: /envs/pytorch/bin/python /envs/pytorch/lib/python3.6/site-packages/pip/_vendor/pep517/in_process/_in_process.py build_wheel /tmp/tmph38jcm_c
       cwd: /tmp/pip-install-fm0wn7au/opencv-python_5a35efbfe61044e6b6c21b37eabf123f
  Complete output (80 lines):
  ...
  ----------------------------------------
  ERROR: Failed building wheel for opencv-python
Failed to build opencv-python
ERROR: Could not build wheels for opencv-python, which is required to install pyproject.toml-based projects

opencv-python的包使用默认pip安装时，会下载最新版本，与当前环境不匹配，需要手动下载当前python版本对应的opencv-python对应版本。

pip install -i https://pypi.douban.com/simple/ pip install opencv-python==4.3.0.38

2、安装opencv-python-headless报错

(pytorch) [root@glnode02 yolov5]# pip install -r requirements.txt -i https://pypi.douban.com/simple
Requirement already satisfied: numpy>=1.13.3 in /envs/pytorch/lib64/python3.6/site-packages (from opencv-python-headless) (1.19.5)
Building wheels for collected packages: opencv-python-headless
  Building wheel for opencv-python-headless (pyproject.toml) ... error
  ERROR: Command errored out with exit status 1:
   command: /envs/pytorch/bin/python /envs/pytorch/lib/python3.6/site-packages/pip/_vendor/pep517/in_process/_in_process.py build_wheel /tmp/tmpum67u7q1
       cwd: /tmp/pip-install-xajga76g/opencv-python-headless_367e20c10e364781b1076237ffd6527d
  Complete output (80 lines):
  ...
  ERROR: Failed building wheel for opencv-python-headless
Failed to build opencv-python-headless
ERROR: Could not build wheels for opencv-python-headless, which is required to install pyproject.toml-based projects

使用pip下载 opencv-python-headless 失败，尝试使用.whl安装包手动安装。
在官网下载对应python版本的opencv-python-headless包：
https://pypi.org/project/opencv-python-headless/3.4.18.65/#files，注意python3.6、linux、x86_64。

pip install opencv_python_headless-3.4.18.65-cp36-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl 

安装完成后，继续下载配置yolov5环境。

待更新。。。