dohbless/yolov5_record-

based on tf/pytorch

yolov5_record

- based on tf/pytorch

驱动版本过低重新安装23333（11.14）

 - https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html（官网查对应驱动版本）
 - 在软件更新里直接安装450版
 - 果然....开始报错NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.
 - 前面加个sudo即可？？？

unload original CUDA 10.0

- cd /usr/local/cuda-10.0/bin/
- sudo ./uninstall_cuda_10.0.pl
- sudo rm -rf /usr/local/cuda-8.0/

check 对应cuda版本

- https://pytorch.org/

官网下载对应版本CUDA

- https://developer.nvidia.com/cuda-10.1-download-archive-update2?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1804&target_type=deblocal
- （实测deb版本最稳定）
- 安装报错 https://blog.csdn.net/wanttifa/article/details/107548931
-配置环境变量
gedit ~/.bashrc
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.1/lib64
export PATH=$PATH:/usr/local/cuda-10.1/bin
export CUDA_HOME=/usr/local/cuda-10.1
source ~/.bashrc更新

sudo /usr/local/cuda-9.1/bin/uninstall_cuda_9.1.pl
- nivdia驱动重新配置
sudo dkms install -m nvidia -v 418.87
-修复环境变量问题
https://blog.csdn.net/qingfenglu/article/details/80388522
https://blog.csdn.net/qq_15265729/article/details/85201786

- cudnn安装
https://blog.csdn.net/wanzhen4330/article/details/81704474
sudo chmod +r libcudnn.so.8.0.2
sudo ln -sf libcudnn.so.8.0.2 libcudnn.so.8  
sudo ln -sf libcudnn.so.8 libcudnn.so     
sudo ldconfig

 sudo ldconfig /usr/local/cuda-10.1/lib64

 tar -xzvf cudnn-10.1-linux-x64-v8.0.2.39.tgz
 https://blog.csdn.net/IT_zxl001/article/details/89350373#commentBox
 (据说cudnn版本8之后无法按原方式显示版本）

所以为啥要卸载重安23333....就很气<>

 - https://blog.csdn.net/qq_38163755/article/details/88353898 (适合deb版本CUDA10.1)
 https://blog.csdn.net/mbytes/article/details/102901486?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param#cuda_75(run版本10.2)
 - chmod a+x cuda_10.2.89_440.33.01_linux.run
 - sudo sh cuda_10.2.89_440.33.01_linux.run
 - sudo gedit .bashrc 等
 - 再次卸载CUDNN
 - 再次安装CUDNN
 tar -xzvf cudnn-10.2-linux-x64-v8.0.3.33.tgz
 https://blog.csdn.net/wanzhen4330/article/details/81704474
 dohbless@dohbless-G3-3579:~/下载/cuda$ sudo cp lib64/lib* /usr/local/cuda/lib64/
(不完整展示)
 sudo chmod +r libcudnn.so.8.0.3
 sudo ln -sf libcudnn.so.8.0.3 libcudnn.so.8
 sudo ln -sf libcudnn.so.8 libcudnn.so     
 sudo ldconfig
 cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

不小心删除了...再次重安

- 先卸载
 sudo /usr/local/cuda-10.2/bin/cuda-uninstaller
 
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.2/lib64
export PATH=$PATH:/usr/local/cuda-10.2/bin
export CUDA_HOME=/usr/local/cuda-10.2

开始重新安装10.2

(这次是提前下好的.run文件啦)

测试之前安好的pytorch

lirui@lirui:~/下载$ python
Python 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 18:10:19) 
[GCC 7.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> 
木有报错嘻嘻

pip 安装pytorch

pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -i  https://pypi.tuna.tsinghua.edu.cn/simple
pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -i http://pypi.douban.com/simple/
https://download.pytorch.org/whl/torch_stable.html
pip --default-timeout=100 install torch torchvision -i  https://pypi.tuna.tsinghua.edu.cn/simple
https://blog.csdn.net/qq_15192373/article/details/104244743 简单直白

realsense相机驱动安装

https://github.com/IntelRealSense/librealsense/blob/master/doc/distribution_linux.md
shell执行命令
realsense-viewer

ros系统安装

- 第一个：sudo sh -c 'echo "deb http://packages.ros.org/ros/ubuntu $(lsb_release -sc) main" > /etc/apt/sources.list.d/ros-latest.list'
- 第二个：sudo apt-key adv --keyserver 'hkp://keyserver.ubuntu.com:80' --recv-key C1CF6E31E6BADE8868B172B4F42ED6FBAB17C654
- 第三个：
sudo apt update
sudo apt install ros-melodic-desktop-full
- 第四个：
sudo rosdep init
rosdep update
如果失败：#打开hosts文件
sudo gedit /etc/hosts
#在文件末尾添加
151.101.84.133 raw.githubusercontent.com
#保存后退出再尝试
来自：https://blog.csdn.net/u013468614/article/details/102917569
- 第五个：
echo "source /opt/ros/melodic/setup.bash" >> ~/.bashrc
source ~/.bashrc
- 最后一个：
sudo apt install python-rosinstall python-rosinstall-generator python-wstool build-essential

apt和dpkg区别

- 两者的区别是dpkg绕过apt包管理数据库对软件包进行操作，所以你用dpkg安装过的软件包用apt可以再安装一遍，系统不知道之前安装过了，将会覆盖之前dpkg的安装。
- 1、dpkg是用来安装.deb文件,但不会解决模块的依赖关系,且不会关心ubuntu的软件仓库内的软件,可以用于安装本地的deb文件。
- 2、apt会解决和安装模块的依赖问题,并会咨询软件仓库, 但不会安装本地的deb文件, apt是建立在dpkg之上的软件管理工具。

ROS+python3+cv_brige

- https://zhuanlan.zhihu.com/p/77682229
- https://www.jianshu.com/p/2f86607c98d1
- https://www.jianshu.com/go-wild?ac=2&url=https%3A%2F%2Fblog.csdn.net%2Flightnateriver%2Farticle%2Fdetails%2F97794261
- https://www.it610.com/article/1279274013260005376.htm

出现错误ModuleNotFoundError: No module named 'defusedxml'

- sudo ln -s /usr/bin/python2.7 /usr/bin/python
- https://blog.csdn.net/wangguchao/article/details/82151372
- https://blog.csdn.net/java0fu/article/details/106081845?utm_medium=distribute.pc_relevant.none-task-blog-title-3&spm=1001.2101.3001.4242

启动ros节点(ctrl+alt+t打开终端 /rosrun运行节点)

roscore
rosrun turtlesim turtlesim_node
rosrun turtlesim turtle_teleop_key

ros学习(tab补全）

https://blog.csdn.net/weixinhum/article/details/83026236
rqt_graph
rosnode list    rosnode info /turtlesim    
rostopic list
rostopic pub /turtel/cmd_vel
- rostopic pub -r 10 /turtle1/cmd_vel geometry_msgs/Twist  "linear  （话题相关）
- rosservice ....(服务相关)
- rosbag 记录话题工具
   rosbag record -a -O cmd_record
   rosbag play ....

ros创建工作空间和工作包

mkdir catkin_ws
cd catkin_ws/
mkdir src
cd src/
catkin_init_workspace
(编译)
cd ..
catkin_make
catkin_make install
(创建工作包)
cd src/
catkin_create_pkg test_pkg std_msgs rospy roscpp
(编译工作包)
cd ..
catkin_make
source devel/setup.bash
(报错)(make: *** 没有规则可以创建“/usr/lib/x86_64-linux-gnu/libGL.so”需要的目标“XXX”）
 sudo ln -s /usr/lib/x86_64-linux-gnu/libGL.so.1.0.0 /usr/lib/x86_64-linux-gnu/libGL.so

ros编译.zip工作包

unzip 2020RC-main.zip
https://blog.csdn.net/weixinhum/article/details/83026236
source ~/RC_record/ROS/devel/setup.bash
source devel/setup.bash
roslaunch main program.launch

在ROS使用RealSense SR300相机

- https://github.com/IntelRealSense/realsense-ros
export ROS_VER=melodic 
sudo apt-get install ros-$ROS_VER-realsense2-camera
sudo apt-get install ros-$ROS_VER-realsense2-description

vscode下配置开发ROS项目

- https://blog.csdn.net/weixin_35695879/article/details/85254422
- 640*480
- 49
- x

.pt转engine

将解释器指向ppython3
dohbless@dohbless-G3-3579:~$ echo alias python=python3 >> ~/.bashrc
dohbless@dohbless-G3-3579:~$ source ~/.bashrc
dohbless@dohbless-G3-3579:~$ python --version
Python 3.6.9
https://github.com/wang-xinyu/tensorrtx/tree/master/yolov5
将best.pt放到yolov5/weights下，并修改为yolov5x.wits
报错No module named 'tqdm'
sudo easy_install pip
pip install tqdm

sudo ./yolov5 -x
报错找不到动态链接库
error while loading shared libraries: libmyelin.so.1: cannot open shared object file: No such file or directory
find / -name libmyelin.so.1
报错CUDA初始化错误，然后测试发现torch.cuda.is_available(） 是False
说是cuda toolkit版本号 和 显卡驱动对应不上
https://lawson-t.blog.csdn.net/article/details/105163179?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param

tensorRt配置

列一下版本号

Ubuntu 18.04
Cuda：10.2
Cudnn：8.0.3
TensorRT：7.1.3.4

配置过程

（https://blog.csdn.net/qq_19707521/article/details/105413411?utm_medium=distribute.pc_aggpage_search_result.none-task-blog-2~all~sobaiduend~default-1-105413411.nonecase&utm_term=ubuntu%20%E9%85%8D%E7%BD%AEtensorrt%E7%8E%AF%E5%A2%83&spm=1000.2123.3001.4430）
-  解压tar包
    tar -zxvf  TensorRT-7.1.3.4.Ubuntu-18.04.x86_64-gnu.cuda-10.2.cudnn8.0.tar.gz
- 环境变量设置
    gedit ~/.bashrc
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/media/dohbless/R/TensorRT-7.1.3.4/lib
    export TENSORRT_ROOT=/media/dohbless/R/TensorRT-7.1.3.4
    source ~/.bashrc
- 安装TensorRT的python接口
    cd TensorRT-7.x.x.x/python
    pip install tensorrt-7.1.3.4-cp36-none-linux_x86_64.whl
 - 安装UFF（Tensorflow所使用的）
    cd TensorRT-7.x.x.x/uff
    pip install uff-0.6.9-py2.py3-none-any.whl
 - 安装graphsurgeon
    cd TensorRT-7.x.x.x/graphsurgeon
    pip install graphsurgeon-0.4.1-py2.py3-none-any.whl
    (https://blog.csdn.net/zong596568821xp/article/details/86077553?utm_medium=distribute.pc_relevant.none-task-blog-utm_term-7&spm=1001.2101.3001.4242)

重新安装opencv4

https://blog.csdn.net/weixin_41921520/article/details/97927633