Install ubuntu Jetson TX2 + Tensorflow

Host软件的安装

主机需要安装的软件列表有Cuda / Tensorrt / Tensorflow GPU/OpenCv/cmake。安装以上软件需要有NvidiaGPU的支持，如果没有。可以安装没有GPU的版本

1.安装Cuda9.0

可以参考这个英文链接

1. 更新显卡驱动

   #自动更新显卡驱动，更新后可能会出现只有独显插口的显示屏才能亮
   sudo add-apt-repository ppa:graphics-drivers/ppa
   sudo apt update 
   

   检查显卡驱动是否更新
   打开ubuntu右上角设置-> about this computer
   如果出现了显卡型号说明显卡已经更新
   如果没有更新
   打开系统设置->软件更新->附加驱动->更新为Nvdia的驱动

2. 安装 cuda，先到cuda官网下载cuda9.0的安装包。由于jetson-x2软件的安装包的cuda型号为9.0所以为保证一致性，也要安装cuda9.0

   sudo dpkg -i cuda-      #小技巧终端输入命令时可以用tab补全文件名
   sudo apt-key add /var/cuda- "autofill with tab" /7fa2af80.pub 
   sudo apt-get update
   sudo apt-get install cuda
   

3. link cuda to bash. //很多软件都需要这个步骤

   sudo nano ~/.bashrc
   

   打开vi编辑器(只能用键盘操作)，用键盘移动到最后一行，加入以下两行代码

   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}}
   

nvcc -V //查看cuda版本

2.安装 tensorrt 4.0

参考这个链接

1. 下载nv-tensorrt-repo-ubuntu1604-cuda9.0-ga-trt4.0.1.6-20180612_1-1_amd64.deb 的deb安装包
2. 终端输入以下代码

   sudo dpkg -i nv-tensorrt "autofill with tab" .deb
   sudo apt-key add /var/nv-tensorrt "autofill with tab" /7fa2af80.pub 
   sudo apt-get update  //不更新可能会报错
   sudo apt-get install tensorrt
   

3.安装pycuda for tensorrt

没有安装pip可以参考终端报错安装pip ·教程

pip install 'pycuda>=2017.1.1'

4.安装bazel 0.15.0

install bazel 0.15.0 with binary installer，
参考教程

sudo apt-get install pkg-config zip g++ zlib1g-dev unzip python  
chmod +x bazel-<version>-installer-linux-x86_64.sh #前提已经下载安装包到本地，改变文件运行权限
./bazel-<version>-installer-linux-x86_64.sh --user
cd ~
gedit .bashrc #最后两行加入下面两句话
source /home/lilanluo/.bazel/bin/bazel-complete.bash
export PATH="$PATH:$HOME/bin"
#更新
source ~/.bashrc

5. 安装cmake

cmake用于编译执行C++软件，安装openCv也需要cmake
参考官方教程

1. 下载 cmake-version.zip 本教程使用的是cmake-3.13.4

   unzip  cmake-version.zip  
   cd cmake-folder  		
   ./bootstrap
   make
   make install
   

6. 安装支持tensorrt的tensorflow 的GPU版本

参考教程

1. 安装支持包

   sudo apt-get install git python-dev python3-dev build-essential swig libcurl3-dev libcupti-dev golang libjpeg-turbo8-dev make tmux htop cmake zlib1g-dev libjpeg-dev xvfb libav-tools xorg-dev python-opengl libboost-all-dev libsdl2-dev apt-transport-https ca-certificates curl software-properties-common coreutils mercurial libav-tools libsdl-image1.2-dev libsdl-mixer1.2-dev libsdl-ttf2.0-dev libsmpeg-dev libsdl1.2-dev libportmidi-dev libswscale-dev libavformat-dev libavcodec-dev libtiff5-dev libx11-6 libx11-dev fluid-soundfont-gm timgm6mb-soundfont xfonts-base xfonts-100dpi xfonts-75dpi xfonts-cyrillic fontconfig fonts-freefont-ttf wget unzip git nasm tar libbz2-dev libgtk2.0-dev libfluidsynth-dev libgme-dev libopenal-dev timidity libwildmidi-dev python-pip python3-pip python-wheel python3-wheel python-virtualenv
   #如果中途报错error则
   sudo apt-get update //更新资源包后再执行一遍上述命令
   
   

2. 安装numpy等支持包

   pip3 install -U --user keras_applications==1.0.6 --no-deps
   pip3 install -U --user keras_preprocessing==1.0.5 --no-deps
   pip3 install numpy
   

3. 用git 下载源码

   cd ~ 
   git clone --single-branch -b v1.10.0 https://github.com/tensorflow/tensorflow
   cd ~/tensorflow
   git checkout v1.10.0
   ./configure
   

   期间终端的应答参考教程即可，基本都是enter就好
   
   ./configure需要的步骤主要是

	1. python  -/usr/bin/python3
	2. cuda9.0 -yescuda9.0 -yes
	3. cudadnn -7.1
	4. NCCL -1.3
	5. note:: 如果cuDNN报错，原因可能是tensorrt没装好，可以再运行一次以下两个步骤

6. bazel build

   bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
   #如果报错，尝试bazel clean 后再执行一遍上述步骤
   bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg/
   pip3 install /tmp/tensorflow_pkg/tensorflow-1.10.0-cp35-cp35m-linux_x86_64.whl
   

8. Install Tensorflow cpp for build with cmake

参考教程

bazel build //tensorflow:libtensorflow_cc.so \
    && sudo mkdir -p                /usr/local/include/google/tensorflow \
    && sudo cp -r bazel-genfiles/*  /usr/local/include/google/tensorflow/ \
    && sudo cp -r tensorflow        /usr/local/include/google/tensorflow/ \
    && sudo find                    /usr/local/include/google/tensorflow -type f  ! -name "*.h" -delete \
    && sudo cp -r third_party       /usr/local/include/google/tensorflow/ \
    && sudo cp bazel-bin/tensorflow/libtensorflow_cc.so /usr/local/lib \
    && sudo cp bazel-bin/tensorflow/libtensorflow_framework.so /usr/local/lib \
    && cd ..

9. 安装Install protobuf 3.6.0

git clone https://github.com/google/protobuf \
    && cd protobuf \
    && git checkout v3.6.0 \   ###注意一定要3.6的版本
    && ./autogen.sh \ #若提示.autoreconf: not found ，则执行
    #sudo apt-get install autoconf automake libtool
    #再执行一遍上述程序 
    && ./configure --prefix=/usr \
    && sudo make install \
    && cd ..

10. Install Eigen3 library

一. Eigen3库可能已经在 /usr/include/.安装了，如果已经有一个版本了，则执行以下步骤：
这个版本可能是错误的，所以需要被tensorflow路经：/tensorflow/bazel-tensorflow/external/eigen_archive 的正确版本替代。方法：

1. 将/tensorflow/bazel-tensorflow/external/eigen_archive 文件复制到 cd/home/lukas and rename it manually to "eigen3"
2. 删除/usr/include的错误版本

   sudo rm -r /usr/include/eigen3
   

3. 复制正确的版本

   sudo cp -r /home/lukas/eigen3 /usr/include 
   

二. 如果eigen3不存在(判断是否存在可以使用)

1. 将/tensorflow/bazel-tensorflow/external/eigen_archive 文件复制到 cd/home/lukas and rename it manually to "eigen3"
2. 复制正确的版本

   sudo cp -r /home/lukas/eigen3 /usr/include 
   

11. how to install openCv3.4

install openCv3.4

12. how to remove oldOpenCvVersion 没用环境会崩掉

reinstall openCv

solve protobuf version Problem

remove old veision

13.Compile cmake project: jetson-workspace.

cd jetson-workspace
cmake CMakeLists.cmake
make
RUN DEMO APPLICATIONS

ATTENTION: TO RUN camera_test with BuildIn Camera install OPENCV3.4.0 with GStreamer 1.0 Support!
src: https://www.jetsonhacks.com/2018/05/28/build-opencv-3-4-with-cuda-on-nvidia-jetson-tx2/

TX2环境搭建

1. 刷机

刷机教程
刷机后板子会装好CUDA / TENSORRT

2. 安装GPU版本的TensorFlow

bash install_tensorflow_on_jetson.sh
bash BuildTensorflow.sh -b v1.10.1 -p python3
./InstallandClean.sh

3. 移动bazel

sudo cp /home/nvidia/JetsonTFBuild/bazel /usr/local/bin

Install Tensorflow cpp for build with cmak

cd JetsonTFBuild/tensorflow

bazel build //tensorflow:libtensorflow_cc.so \
    && sudo mkdir -p                /usr/local/include/google/tensorflow \
    && sudo cp -r bazel-genfiles/*  /usr/local/include/google/tensorflow/ \
    && sudo cp -r tensorflow        /usr/local/include/google/tensorflow/ \
    && sudo find                    /usr/local/include/google/tensorflow -type f  ! -name "*.h" -delete \
    && sudo cp -r third_party       /usr/local/include/google/tensorflow/ \
    && sudo cp bazel-bin/tensorflow/libtensorflow_cc.so /usr/local/lib \
    && sudo cp bazel-bin/tensorflow/libtensorflow_framework.so /usr/local/lib \
    && cd ..

4. 安装Cmake3.13.4

下载安装包，解压， 以.tar.gz结尾的安装包为源码，然后可以用下面的命令在TX2进行编译:

./bootstrap
  make
  make install

5. 安装protobuf 3.6.0```python

git clone https://github.com/google/protobuf \
    && cd protobuf \
    && git checkout v3.6.0 \
    && ./autogen.sh \
    && ./configure --prefix=/usr \
    && sudo make install \
    && cd ..

6. 安装Eigen3lib

• 参考host机改路径即可

1. 考虑到可能在/usr/include/已经有eigen3的了。但是这个 版本需要被/tensorflow/bazel-tensorflow/external/eigen_archive替换

• 从 tensorflow/bazel-tensorflow/external/eigen_archive 复制到 /home/Nvidia并且重命名为eigen3

2. 删除/usr/include的错误版本

   sudo rm -r /usr/include/eigen3
   

3. 复制正确的版本

   sudo cp -r /home/Nvidia/eigen3 /usr/include 
   

7. 安装openCV3.4.0

参考host机

8. 编译执行

参考host机子