Skip to content

Latest commit

 

History

History
736 lines (580 loc) · 28.6 KB

单机深度学习环境配置教程.md

File metadata and controls

736 lines (580 loc) · 28.6 KB

DBRG离线安装Ubuntu16.04 NVIDIA驱动 CUDA9.0 CUDNN7.0 anaconda TensorFlow-GPU pycharm opencv-python opencv-contrib-python pytorch clion qt5 OpenCV3.3.1教程

1、格式化原Ubuntu分区

https://jingyan.baidu.com/article/295430f13ed7d80c7e005088.html

2、重装Ubuntu16.04

下载地址: http://mirrors.aliyun.com/ubuntu-releases/16.04/ubuntu-16.04.5-desktop-amd64.iso 参考博客:https://blog.csdn.net/weixin_38233274/article/details/80237572

(1) 将ubuntu-16.04.4-desktop-amd64.iso放到C盘根目录,镜像文件里面有个casper文件夹,将文件vmlinuz 、initrd也拷贝到C盘根目录下。

(2) 运行EasyBCD,“添加新条目”->“NeoGrub”->“安装”。

(3) 配置->编辑menu.lst文件

(4) title Install Ubuntu

root (hd0,0) kernel (hd0,0)/vmlinuz boot=casper iso-scan/filename=/ubuntu-16.04.2-desktop-amd64.iso ro quiet splash locale=zh_CN.UTF-8 initrd (hd0,0)/initrd

(5) 重启(选择NeoGrub)

(6) 在安装之前打开终端Ctrl+Alt+T,输入sudo umount -l /isodevice,注意空格,可多执行一次,以确保将挂载的镜像移除,否则将无法进行安装。

(7) 您已安装的多个操作系统->其他选项

(8) 运行ubuntu安装程序安装Ubuntu16.04 LTS,交换空间一般跟内存条大小差不多就可以了,/和/home平分各100G差不多,最下面的挂载选在/所在的分区,当Windows系统重装时,就不会影响Ubuntu系统了

(9) 安装完成后重启直接进入Windows,运行EasyBCD,“添加新条目”->“NeoGrub”->“删除”,删除ubuntu的安装引导。

(10) EasyBCD,“添加新条目”->“Linux/BSD”。类型选择 Grub2,名称可自定,驱动器选择/所在的分区。点击“添加条目”即可。

(11) 重启即可。删除安装引导选项。EasyBCD软件,进入一开始配置文件的那个位置,点击 remove 即可 ,重新启动就不会有引导安装的选项了。

3、配置固定IP

(1) windows系统下查看自己的IP

(2) Ubuntu下进行网络设置

4、更新源(如果我们的16.04内网源好使的了的话)

教程参考192.168.2.68/ubuntu/mannual.html(如果连不上就是不听话没有配置固定IP)

(1) cd (sources.list位置)

(2) sudo cp sources.list /etc/apt/sources.list

(3) sudo apt-get update

5、安装NVIDIA显卡驱动

下载地址:https://www.nvidia.cn/Download/index.aspx?lang=cn

参考博客:https://blog.csdn.net/xx_katherine/article/details/77754179

(1) 卸载原有驱动sudo apt-get purge nvidia*

(2) 禁用nouveau,创建blacklist-nouveau.conf

sudo vim /etc/modprobe.d/blacklist-nouveau.conf

编辑内容为:

blacklist nouveau
options nouveau modeset=0

(3)更新后重启系统

sudo update-initramfs –u

(4)关闭图形化界面

sudo service lightdm stop

(5)ctrl+alt+f1进入tty1命令行模式安装驱动

cd (驱动位置)
sudo sh ./NVIDIA*.run

(6)安装完成后重启图像化界面

sudo service lightdm start

(7)验证NVIDIA安装成功,成功打印出显卡信息

nvidia-smi

6、安装CUDA9.0

首先我要说一说为什么要安装9.0:

https://stackoverflow.com/questions/50442076/install-gpu-version-tensorflow-with-older-version-cuda-and-cudnn

历史经验告诉我们,我们实验需要TensorFlow-GPU>1.7.0,这就需要CUDA9.0+CUDNN7.0以上的配置(要对应);而cuda9.0没有Ubuntu14的版本。如果你安装的是Ubuntu14.04或者其他低于Ubuntu16.04的版本,然后发现你要使用TensorFlow-GPU1.7.0以上版本的功能,那就可以休息一天,重新在装一遍,这就是为什么有此一文。

下载地址:

https://developer.nvidia.com/cuda-90-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=runfilelocal

参考博客:https://blog.csdn.net/qlulibin/article/details/78714596

(1) 关闭图形化界面,ctrl+alt+f1进入tty1命令行模式安装驱动

(2) 进入run文件位置,执行如下命令,一直回车看完文档

sudo sh cuda_9.0.176_384.81_linux.run

(3) 根据提示输入,默认路径即可

(4) 进入图形化界面配置环境变量,运行如下命令打开profile文件

sudo gedit  /etc/profile

(5) 打开文件后在文件末尾添加路径,也就是安装目录,命令如下:(如果重启后报错,把这两句命令放在.bashrc中,参见cudnn安装报错解决办法)

export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

(6) 保存,然后重启电脑

sudo reboot

(7) 测试CUDA的Samples例子

cd  /usr/local/cuda-9.0/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery

(8) PASS:成功

(9) 安装补丁

7、安装Cudnn

下载地址:https://developer.nvidia.com/rdp/cudnn-download

参考博客:https://www.jianshu.com/p/69a10d0a24b9

验证cudnn正确安装:

https://blog.csdn.net/u014561933/article/details/79968539#4%E9%AA%8C%E8%AF%81

报错:参考博客:https://blog.csdn.net/mumodm/article/details/79502848 (1) 根据如下命令

cd ~
 sudo tar xvf cudnn-8.0-linux-x64-v5.1.tgz
 cd cuda/include
 sudo cp *.h /usr/local/include/
 cd ../lib64
 sudo cp lib* /usr/local/lib/
 cd /usr/local/lib# sudo chmod +r libcudnn.so.5.1.5
 sudo ln -sf libcudnn.so.7.2.1 libcudnn.so.7
 sudo ln -sf libcudnn.so.7 libcudnn.so
 sudo ldconfig

(2)验证是否正确安装

验证包:http://og9m6v6ow.bkt.clouddn.com/cudnn_samples_v7.tar.gz

解压到可写的文件夹下,进入

cd  cudnn_samples_v7/mnistCUDNN

(3)编译

make clean && make

(4)运行mnistCUDNN样例

 ./mnistCUDNN

(5)如果输出:Test passed!说明安装完成

(6)如果过程中报错,大部分情况下是环境没有配好

Error: libcudart.so.9.0: cannot open shared object file: No such file or directory
// 或者
Error: libcusolver.so.9.0: cannot open shared object file: No such file or direcctory
// 或者
Error: libcublas.so.9.0: cannot open shared object file: No such file or directory

参考博客:https://blog.csdn.net/mumodm/article/details/79502848

①第一种可靠的解决方法:

cd ~
sudo vi .bashrc
// 下滑到文件末,添加以下内容
export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/cuda/lib64
// 刷新.bashrc
source .bashrc
// 以上解法是对生成了软连接的情况;如果没有生成软连接,则把以上的cuda改为cuda-9.0

②如果添加好了环境,还是出现同样的报错,则可以尝试以下解法:

cd ~
sudo cp /usr/local/cuda-9.0/lib64/libcudart.so.8.0 /usr/local/lib/libcudart.so.9.0 && sudo ldconfig
cp /usr/local/cuda-9.0/lib64/libcublas.so.9.0 /usr/local/lib/libcublas.so.9.0 && sudo ldconfig
cp /usr/local/cuda-9.0/lib64/libcurand.so.9.0 /usr/local/lib/libcurand.so.9.0 && sudo ldconfig

报哪个错就改哪个

③一般情况下,以上两种解法可以搞定问题的;如果还是报错libcusolver.so.9.0不存在,下面是算是一种解法:

sudo ldconfig /usr/local/cuda/lib64

8、安装anaconda(自带python3.6)

下载地址:https://repo.anaconda.com/archive/Anaconda3-5.2.0-Linux-x86_64.sh

参考博客:https://blog.csdn.net/xiaerwoailuo/article/details/70054429

(1)在命令行用python和python3命令查看python版本

Ubuntu16自带的是python2.7和python3.6,安装的

(2)进入Anaconda3-5.2.0-Linux-x86_64.sh文件位置,然后执行

bash Anaconda3-5.2.0-Linux-x86_64.sh

(3)一路回车/yes,会自动配置好环境变量,重启终端才会生效。重启后输入python,提示python 3.6.5 anaconda……说明安装完成

(4)通过import scipy验证是否安装成功

9、安装TensorFlow-GPU

下载地址:https://pypi.org/project/tensorflow-gpu/#files

参考博客:https://blog.csdn.net/taoqick/article/details/79171199

(1)进入文件路径

pip install tensorflow_gpu-1.10.1-cp36-cp36m-manylinux1_x86_64.whl

(2)安装过程中会报错,是因为离线安装缺少依赖包,踩过坑的会把包留着(这就是为什么上一步要安装anaconda,一方面anaconda方便管理python版本,另一方面就是会自动安装很多包,所以这一步也就几个文件需要自己手动安装),但系统不会自动安装压缩文件,pip install ……重复直到TensorFlow-GPU安装成功即可

(3)验证是否安装成功

python
>>>import tensorflow as tf
>>>a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
>>>b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
>>>c = tf.matmul(a, b)# Creates a session with log_device_placement set to True.
>>>sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))# Runs the op.
>>>print(sess.run(c))

(4)输出结果证明安装成功

10、安装pycharm

下载地址: https://download.jetbrains.8686c.com/python/pycharm-community-2018.2.3.tar.gz

参考博客:https://blog.csdn.net/qq_38786209/article/details/78309191?readlog https://blog.csdn.net/sinat_35257860/article/details/72737399

(1)进入文件路径 tar -xvzf pycharm-community-2018.2.3.tar.gz

(2)进入解压路径,运行

cd (解压文件路径)pycharm-community-2018.2.3/bin sh pycharm.sh

(3)Pycharm启动方法:

参考博客:https://blog.csdn.net/sinat_35257860/article/details/72737399

a)sh pycharm.sh

b)https://blog.csdn.net/tmosk/article/details/72852330

cd /usr/share/applications/
sudo vim Pycharm.desktop

这里必须得用root权限sudo才能写入,然后在文件中写入以下内容。

 [Desktop Entry]
Type = Application     
Name = Pycharm
GenericName = Pycharm
Comment = Pycharm:The Python IDE
Exec = sh /home/lxq/Downloads/pycharm/bin/pycharm.sh
Icon = /home/lxq/Downloads/pycharm/bin/pycharm.png
Terminal = pycharm
Categories = Pycharm;

c)在pycharm工具里选择创建图标:Tools -> create desktop entry...(亲测这个最方便)

(4)配置编译环境file->settings->小齿轮->add->选择/usr/local/anaconda3bin/python3.6(总之是python3.6,选择所有项目都使用这个编译器。因为TensorFlow是这个版本的,没有他用其他编译器也可以)

(5)新建文件测试,成没成功你知道的

import tensorflow as tf
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))# Runs the op.
print(sess.run(c))

11、Anaconda下安装python版本的opencv-python和opencv-contrib-python

下载地址:https://pypi.org/project/opencv-python/ https://pypi.org/project/opencv-contrib-python/

参考博客:同上 进入这两个文件位置,在终端中输入两句命令:

pip install opencv_python-3.4.3.18-cp36-cp36m-manylinux1_x86_64.whl
pip install opencv_contrib_python-3.4.3.18-cp36-cp36m-manylinux1_x86_64.whl

OK,测试一下import cv2,成功

12、安装pytorch

下载地址: Pytorch:https://pytorch.org/previous-versions/

torchvision 0.2.1:https://pypi.org/project/torchvision/#files

参考博客:https://blog.csdn.net/red_stone1/article/details/78727096

(1)进入PyTorch的下载目录,使用pip命令安装:

pip install torch-0.4.0-cp36-cp36m-linux_x86_64.whl

(2)在pypi下载,然后安装torchvision,可直接使用pip命令安装:

pip install torchvision

(3)测试,进入python环境

import torch
import torchvision
print(torch.cuda.is_available())#输出true
exit()

13、安装clion

下载地址:https://download.jetbrains.8686c.com/cpp/CLion-2018.2.3.tar.gz

参考博客:https://blog.csdn.net/u010925447/article/details/73251780

(1) tar -zxvf CLion-2016.2.2.tar.gz

(2) cd clion-2016.2.2/bin/

(3) ./clion.sh

(4) 验证码http://idea.lanyus.com/

(5) 注意新建工程测试的时候要把对应的CMakeList.txt中cmake的版本改成自己的!

14、安装QT

下载地址:https://pan.baidu.com/s/1o7H1y2I

参考博客:https://blog.csdn.net/lql0716/article/details/54564721

(1)将下载的安装文件qt-opensource-linux-x64-5.7.1.run拷贝到home/用户目录,如/home/user (2)如果qt-opensource-linux-x64-5.7.1.run的属性中拥有者没有运行权限,则可用chmod命令添加执行权限: (3)chmod u+x qt-opensource-linux-x64-5.7.1.run (4)在终端执行:

./ qt-opensource-linux-x64-5.7.1.run

(5)跳出安装界面,一直点击下一步,直到安装完成即可。

(6)测试控制台程序

#include <QCoreApplication>
#include <stdio.h>
#include <iostream>
int main(int argc, char *argv[])
{
    QCoreApplication a(argc, argv);
    std::cout<<”hello”<<std::endl;#Ubuntu下printf不好使哦
    return a.exec();
}

15、安装opencv3.3和opencv_contrib

下载地址:https://github.com/opencv/opencv/archive/3.3.1.zip

https://github.com/opencv/opencv_contrib/archive/3.3.1.zip

参考博客:https://www.cnblogs.com/arkenstone/p/6490017.html

https://blog.csdn.net/xiangxianghehe/article/details/78780269

(1)安装依赖包

sudo apt-get install build-essential  
sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev  
sudo apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev  
sudo apt-get install build-essential qt5-default ccache libv4l-dev libavresample-dev  libgphoto2-dev libopenblas-base libopenblas-dev doxygen  openjdk-8-jdk pylint libvtk6-dev
sudo apt-get install pkg-config

(2)解压下载好的包:

unzip opencv-3.3.1.zip
unzip opencv_contrib-3.3.1.zip

(3)解压完后需要将opencv_contrib.zip提取到opencv目录下,同时在该目录下新建一个文件夹build:

cp -r opencv_contrib-3.3.1 opencv-3.3.1  #复制opencv_contrib到opencv目录下
cd opencv-3.3.1
mkdir build   #新建文件夹build

(4) 进入build目录,并且执行cmake生成makefile文件:

cd build  

(5)

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D INSTALL_PYTHON_EXAMPLES=ON -D INSTALL_C_EXAMPLES=ON -D OPENCV_EXTRA_MODULES_PATH=/home/elsie/OPENCV/opencv-3.3.1/opencv_contrib-3.3.1/modules -D WITH_CUDA=ON -D WITH_CUBLAS=ON -D DCUDA_NVCC_FLAGS="-D_FORCE_INLINES" -D CUDA_ARCH_BIN="6.1" -D CUDA_ARCH_PTX="" -D CUDA_FAST_MATH=ON -D WITH_TBB=ON -D WITH_V4L=ON -D WITH_QT=ON -D WITH_GTK=ON -D WITH_OPENGL=ON -D BUILD_EXAMPLES=ON ..

注意:①CUDA_ARCH_BIN="6.1”这个需要去官网确认使用的GPU所对应的版本查看这里

②如果qt未安装可以删去此行;若因为未正确安装qt导致的Qt5Gui报错,可将build内文件全部删除后重新cmake,具体可以参考这里

③OPENCV_EXTRA_MODULES_PATH就是你 opencv_contrib-3.3.1下面的modules目录,请按照自己的实际目录修改地址。

④后面的两点不可省略

(6)生成完毕提示:(没有错误!有坑!)

--   Install path:                  /usr/local
--
--   cvconfig.h is in:        /home/elsie/OPENCV/opencv-3.3.1/opencv_contrib-3.3.1 /build
-- -----------------------------------------------------------------
--
-- Configuring done
-- Generating done
-- Build files have been written to: /home/elsie/OPENCV/opencv-3.3.1/opencv_contrib-3.3.1/modules /build

注意:虽然Configuring done -- Generating done这里仍然会有几个坑影响后面的make

1 过程中需要下载诸如ippicv_2017u3_lnx_intel64_20170822.tgz的东西(在cmake的输出中往上拉,一般都会失败),如果下载失败: 下载地址:https://github.com/opencv/opencv_3rdparty/branches/all 下载的东西名叫opencv_3rdparty-ippicv-master_20170822.zip,解压找到ippicv_2017u3_lnx_intel64_general_20170822.tgz文件,拷贝到某目录,然后把~/opencv-3.3.1/3rdparty/ippicv中的ippicv.cmake文件中的GitHub下载地址修改为自己的本地地址。 注意:网页中说的是修改为files://地址,不需要files://,这是从服务器下载,路径直接写文件路径即可。

2 缺少boostdesc_bgm.i boostdesc_bgm_bi.i boostdesc_bgm_hd.i boostdesc_binboost_064.i boostdesc_binboost_128.i boostdesc_binboost_256.i boostdesc_lbgm.i vgg_generated_120.i vgg_generated_48.i vgg_generated_64.i vgg_generated_80.i等文件。

下载地址:https://download.csdn.net/download/sinat_39805237/10563950 所有文件放到opencv_contrib-3.3.1/modules/xfeatures2d/src中 然后把opencv_contrib-3.3.1/modules/xfeatures2d/cmake文件夹里的download_boostdesc.cmake 和download_vgg.cmake中下载地址那一部分改成……/src那一段。

3 es10_300x300_ssd_iter_140000.caffemodel和tiny-dnn下的v1.0.0a3.tar.gz找不到

下载地址:https://download.csdn.net/download/u010782463/10309793 https://download.csdn.net/download/wjskeepmaking/9824941?web=web 解决方法同①,修改~/ opencv-3.3.1/ opencv_contrib-3.3.1/modules/dnn_modern/cmake里的cmakelist.txt改成本地路径

(7) 在cmake成功之后,就可以在build文件下make了:

make -j8        #8线程编译
make install

(8) 测试

/**
* @概述:采用FAST算子检测特征点,采用SIFT算子对特征点进行特征提取,并使用BruteForce匹配法进行特征点的匹配
* @类和函数:FastFeatureDetector + SiftDescriptorExtractor + BruteForceMatcher
*/


#include<opencv2/opencv.hpp>
#include <opencv2/xfeatures2d.hpp>

using namespace std;
using namespace cv;
using namespace cv::xfeatures2d;

int main(int argc, char** argv)
{
    Mat objImage = imread("1.jpg", IMREAD_COLOR);
    Mat sceneImage = imread("2.jpg", IMREAD_COLOR);
    //-- Step 1: Detect the keypoints using SURF Detector
    int minHessian = 400;
    Ptr<SURF> detector = SURF::create(minHessian);
    std::vector<KeyPoint> obj_keypoint, scene_keypoint;
    detector->detect(objImage, obj_keypoint);
    detector->detect(sceneImage, scene_keypoint);
    //computer the descriptors
    Mat obj_descriptors, scene_descriptors;
    detector->compute(objImage, obj_keypoint, obj_descriptors);
    detector->compute(sceneImage, scene_keypoint, scene_descriptors);
    //use BruteForce to match,and get good_matches
    BFMatcher matcher;
    vector<DMatch> matches;
    matcher.match(obj_descriptors, scene_descriptors, matches);
    sort(matches.begin(), matches.end());  //筛选匹配点
    vector<DMatch> good_matches;
    for (int i = 0; i < min(50, (int)(matches.size()*0.15)); i++) {
        good_matches.push_back(matches[i]);
    }
    //draw matches
    Mat imgMatches;
    drawMatches(objImage, obj_keypoint, sceneImage, scene_keypoint,good_matches, imgMatches);
    //get obj bounding
    vector<Point2f> obj_good_keypoint;
    vector<Point2f> scene_good_keypoint;
    for (int i = 0; i < good_matches.size(); i++) {
        obj_good_keypoint.push_back(obj_keypoint[good_matches[i].queryIdx].pt);
        scene_good_keypoint.push_back(scene_keypoint[good_matches[i].trainIdx].pt);
    }
    vector<Point2f> obj_box(4);
    vector<Point2f> scene_box(4);
    obj_box[0] = Point(0, 0);
    obj_box[1] = Point(objImage.cols, 0);
    obj_box[2] = Point(objImage.cols, objImage.rows);
    obj_box[3] = Point(0, objImage.rows);
    Mat H = findHomography(obj_good_keypoint, scene_good_keypoint, RANSAC); //find the perspective transformation between the source and the destination
    perspectiveTransform(obj_box, scene_box, H);
    line(imgMatches, scene_box[0]+Point2f((float)objImage.cols, 0), scene_box[1] + Point2f((float)objImage.cols, 0), Scalar(0, 255, 0), 2);
    line(imgMatches, scene_box[1] + Point2f((float)objImage.cols, 0), scene_box[2] + Point2f((float)objImage.cols, 0), Scalar(0, 255, 0), 2);
    line(imgMatches, scene_box[2] + Point2f((float)objImage.cols, 0), scene_box[3] + Point2f((float)objImage.cols, 0), Scalar(0, 255, 0), 2);
    line(imgMatches, scene_box[3] + Point2f((float)objImage.cols, 0), scene_box[0] + Point2f((float)objImage.cols, 0), Scalar(0, 255, 0), 2);
    //show the result                                                                   
    imshow("匹配图", imgMatches);
    //save picture file
    imwrite("final.jpg",imgMatches);
    waitKey(0);
    return 0;
}

项目的cmakelist.txt配置如下:

cmake_minimum_required(VERSION 3.5)
project(untitled)
set(CMAKE_CXX_STANDARD 14)
set(SOURCE_FILES main.cpp)
add_executable(untitled main.cpp)
find_package(OpenCV)
include_directories(${OpenCV_INCLUDE_DIRS})
set(CMAKE_CXX_STANDARD 14)
set(SOURCE_FILES main.cpp)
target_link_libraries(untitled ${OpenCV_LIBS})

(9) 链接库共享。编译安装完毕之后,为了让你的链接库被系统共享,让编译器发现,需要执行管理命令ldconfig:

sudo ldconfig -v  

16、 【这是一段失败的旅程,后来我放弃了,不过可以解决的…】安装OPENCV2.4.9

下载地址:http://jaist.dl.sourceforge.net/project/opencvlibrary/opencv-unix/2.4.9/opencv-2.4.9.zip 参考博客:(参考的有点多,主要都列在排错上了)

https://blog.csdn.net/u014527548/article/details/80251046

(1) 解压到任意目录,进入源码目录

unzip opencv-2.4.9.zip
cd opencv-2.4.9

(2) 安装下列依赖

sudo apt-get install build-essential cmake libgtk2.0-dev pkg-config python-dev python-numpy libavcodec-dev libavformat-dev libswscale-dev

sudo apt-get install build-essential libgtk2.0-dev libjpeg-dev libtiff4-dev libjasper-dev libopenexr-dev cmake python-dev python-numpy python-tk libtbb-dev libeigen3-dev yasm libfaac-dev libopencore-amrnb-dev libopencore-amrwb-dev libtheora-dev libvorbis-dev libxvidcore-dev libx264-dev libqt4-dev libqt4-opengl-dev sphinx-common texlive-latex-extra libv4l-dev libdc1394-22-dev libavcodec-dev libavformat-dev libswscale-dev
default-jdk ant libvtk5-qt4-dev

注意:这里可能会报错:

libgtk2.0-dev : 依赖: libgtk2.0-0 (= 2.24.23-0ubuntu1) 但是 2.24.23-0ubuntu1.1 正要被安装
                 依赖: libpango1.0-dev (>= 1.20) 但是它将不会被安装
                 依赖: libcairo2-dev (>= 1.6.4-6.1) 但是它将不会被安装
                 推荐: debhelper 但是它将不会被安装
E: 无法修正错误,因为您要求某些软件包保持现状,就是它们破坏了软件包间的依赖关系。

如果忽略了这个错误继续安装,后面的OpenCV可能不能正常使用,我们要解决这个问题。

方法:

sudo aptitude install libgtk2.0-dev

下面会出来一堆解决方案,都是保留……,然后问是否接受这个解决方案。

这时候要选No!因为出现这个问题的根本原因是安装包A依赖于C的旧版本,而机器上已经存在了C的新版本,此新的版本又是B的依赖,所以就会出现版本的依赖混乱问题。 直到出现了“降级”这样的解决方案,yes。降级完之后重新安装即可。其他类似问题同样可以参考。

也有可能是源的问题,不过,离线安装既然做不到在线更新源,那就酱紫继续吧。

还有,可能会报python-numpy的包依赖错误。这里我是先装好了TensorFlow之前一套,才想起来更新源的。不知道是不是这个原因导致的依赖问题,python-numpy降级以后,记得先测试一下import tensorflow as tf能否正常工作,如果不行的话,再测cuda\cudnn能否运行示例程序。

我这里是可以运行示例程序,但找不到TensorFlow,现在需要重装TensorFlow。步骤如上……

然后保险起见再执行一下

sudo apt-get install build-essential cmake libgtk2.0-dev pkg-config python-dev python-numpy libavcodec-dev libavformat-dev libswscale-dev

(3)进入cmake

cd cmake

(4) cmake编译生成Makefile,安装所有的lib文件都会被安装到/usr/local目录

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local ..  

不报错的人生一点都不完美

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_nppi_LIBRARY (ADVANCED)
linked by target "opencv_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev
linked by target "opencv_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev
linked by target "opencv_test_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev/test
linked by target "opencv_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_test_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_perf_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
……
Please set them or make sure they are set and tested correctly in the CMake files:
CUDA_nppi_LIBRARY (ADVANCED)
linked by target "opencv_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev
linked by target "opencv_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev
linked by target "opencv_test_cudev" in directory D:/Cproject/opencv/opencv/sources/modules/cudev/test
linked by target "opencv_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_test_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_perf_core" in directory D:/Cproject/opencv/opencv/sources/modules/core
linked by target "opencv_test_cudaarithm" in directory

……

为啥呢?满屏都是错

大神告诉我们,原因是cuda9.0不再支持2.0架构

参考博客:https://blog.csdn.net/u014613745/article/details/78310916 https://blog.csdn.net/mystylee/article/details/79035585

https://stackoverflow.com/questions/46584000/cmake-error-variables-are-set-to-notfound

解决方案抄录如下:(注意OpenCV2版本的不执行https://blog.csdn.net/u014613745/article/details/78310916中的第四步,否则会报错;OpenCV3架构的需要执行,否则会报错)

4 找到FindCUDA.cmake文件(opencv-2.4.9下cmake目录),找到行find_cuda_helper_libs(nppi)修改为:

find_cuda_helper_libs(nppial)
find_cuda_helper_libs(nppicc)
find_cuda_helper_libs(nppicom)
  find_cuda_helper_libs(nppidei)
  find_cuda_helper_libs(nppif)
  find_cuda_helper_libs(nppig)
  find_cuda_helper_libs(nppim)
  find_cuda_helper_libs(nppist)
find_cuda_helper_libs(nppisu)
 find_cuda_helper_libs(nppitc)

5 找到行

set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppi_LIBRARY};${CUDA_npps_LIBRARY}")修改为
set(CUDA_npp_LIBRARY "${CUDA_nppc_LIBRARY};${CUDA_nppial_LIBRARY};${CUDA_nppicc_LIBRARY};${CUDA_nppicom_LIBRARY};${CUDA_nppidei_LIBRARY};${CUDA_nppif_LIBRARY};${CUDA_nppig_LIBRARY};${CUDA_nppim_LIBRARY};${CUDA_nppist_LIBRARY};${CUDA_nppisu_LIBRARY};${CUDA_nppitc_LIBRARY};${CUDA_npps_LIBRARY}")

6 找到行

unset(CUDA_nppi_LIBRARY CACHE)修改为
unset(CUDA_nppial_LIBRARY CACHE)
unset(CUDA_nppicc_LIBRARY CACHE)
unset(CUDA_nppicom_LIBRARY CACHE)
unset(CUDA_nppidei_LIBRARY CACHE)
unset(CUDA_nppif_LIBRARY CACHE)
unset(CUDA_nppig_LIBRARY CACHE)
unset(CUDA_nppim_LIBRARY CACHE)
unset(CUDA_nppist_LIBRARY CACHE)
unset(CUDA_nppisu_LIBRARY CACHE)
unset(CUDA_nppitc_LIBRARY CACHE)

7 cuda9中有一个单独的halffloat(cuda_fp16.h)头文件,也应该被包括在opencv的目录里,将头文件cuda_fp16.h添加至 opencv\modules\gpu\include\opencv2\gpu\common.hpp 即在common.hpp中添加

#include <cuda_fp16.h>

8 重新执行

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local ..  

(5) 继续OpenCV的安装,在cmake文件夹下执行以下命令

sudo make install

(6) 还是会报错,不报错的人生不完美。这次的错误长这样:

nvcc fatal   : Unsupported gpu architecture 'compute_11'
CMake Error at cuda_compile_generated_matrix_operations.cu.o.cmake:208 (message):
Error generating/home/elsie/Documents/opencv2.4.9/build/modules/core/CMakeFiles/cuda_compile.dir/__/dynamicuda/src/cuda/./cuda_compile_generated_matrix_operations.cu.o
make[2]: ***
[modules/core/CMakeFiles/cuda_compile.dir/__/dynamicuda/src/cuda/./cuda_compile_generated_matrix_operations.cu.o] Error 1
make[1]: *** [modules/core/CMakeFiles/opencv_core.dir/all] Error 2 make[1]: *** Waiting for unfinished jobs.…

解决一下吧(虽然只写了五个字,可是我卡了半天):

cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D CUDA_GENERATION=Kepler ..

然后就按照上面的解决办法把丢失的文件补进去就好了。

作者:徐樱笑