目录
4.2.2 查看jetpack与tensorflow的对应关系
11 复制 JetBot notebooks 到home目录
JetBot 项目是NVIDIA（英伟达）基于Jetson Nano开发套件而设计的一款开源智能车项目。 它提供了基于视觉的自主避障，物体跟随，路径跟踪以及人脸识别等功能。对于AI初学者来说，Jetbot是一个很好的入门途径。
Nvidia官方提供了完整的硬件和软件列表，参考https://github.com/NVIDIA-AI-IOT/jetbot/wiki， 可以自行组装jetbot小车。但是，自己采购零件还是很麻烦，国内的微雪公司提供了完整的小车方案，可以直接从https://www.waveshare.net/list.html?cat=261购买。
硬件有了，软件环境的搭建也是一个大工程。Nvidia官方提供了Jetbot镜像下载(https://drive.google.com/open?id=1G5nw0o3Q6E08xZM99ZfzQAe7-qAXxzHN)， 但是很不幸，镜像位于google仓库，国内如果没有翻墙软件，根本无法下载. 本文教你如何自己从头构建一个jetbot环境。
主要参考https://github.com/NVIDIA-AI-IOT/jetbot/wiki/Create-SD-Card-Image-From-Scratch和https://blog.csdn.net/liam_dapaitou/article/details/90451808。
1. 烧录Jetson Nano镜像
这一步简单，直接参照Nvidia的官方步骤即可。
https://developer.nvidia.com/embedded/learn/get-started-jetson-nano-devkit#write
有的文章中要求ubuntu系统的用户名和密码需要为jetbot /jetbot， 按照我的实践结果，并不需要，使用自己喜欢的用户名和密码就可以。
2 设置i2c permissions
sudo usermod -aG i2c $USER
3 安装pip3和python依赖
sudo apt-get updatesudo apt install python3-pip python3-pilsudo pip3 install --upgrade numpy
4 安装TensorFlow
这里https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html
给出了官方步骤在jetson nano上安装tensorflow。
4.1 安装依赖
$ sudo apt-get update$ sudo apt-get install libhdf5-serial-dev hdf5-tools libhdf5-dev zlib1g-dev zip libjpeg8-dev liblapack-dev libblas-dev gfortran$ sudo apt-get install python3-pip$ sudo pip3 install -U pip testresources setuptools$ sudo pip3 install -U numpy==1.16.1 future==0.17.1 mock==3.0.5 h5py==2.9.0 keras_preprocessing==1.0.5 keras_applications==1.0.8 gast==0.2.2 futures protobuf pybind11
4.2 确定需要的TensorFlow版本
Jetson Nano镜像一直在更新jetpack的版本，随之使用的tensorflow版本也需要更新。所以首先需要确定自己的Jetpack版本，以及与tensorflow版本的对应关系。
4.2.1 查看jetpack 版本
运行 $ head -n 1 /etc/nv_tegra_release
# R32 (release), REVISION: 4.3, GCID: 21589087, BOARD: t210ref, EABI: aarch64, DATE: Fri Jun 26 04:38:25 UTC 2020
note that, that version is jetpack4.4, not 4.3
4.2.2 查看jetpack与tensorflow的对应关系
下面链接给出了这个对应关系：
我这里选的是最新的2.2.0, container20.07
4.3 下载TensorFlow
下面的网址：https://developer.nvidia.com/embedded/downloads#?search=tensorflow
选择需要的包下载。
4.4 安装TensorFlow
如下命令安装：
$ sudo pip3 install /local/jetbot/tensorflow-2.2.0+nv20.6-cp36-cp36m-linux_aarch64.whl
安装会持续很长时间，尤其在安装 sudo pip3 install grpcio 时，需要耐心等待。
4.5 验证安装是否成功
1) 运行:
$ python3
2) Import TensorFlow:
>>> import tensorflow
如果安装成功，不会报任何错误.
如果报错误
2020-07-30 22:25:36.863526: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.0'; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory
2020-07-30 22:25:36.863829: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.0'; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory
Segmentation fault (core dumped)
这个问题的出现是由于目前版本的Tensorflow还只能支持CUDA10.0，而英伟达的CUDA则是更新到了10.2，要解决这个问题，可以下载更新版本的TensorFlow。
查看CUDA的版本：
/usr/local/cuda-10.2/doc/man/man7/libcudart.so.7
/usr/local/cuda-10.2/targets/aarch64-linux/lib/libcudart.so.10.2.89
/usr/local/cuda-10.2/targets/aarch64-linux/lib/libcudart.so
/usr/local/cuda-10.2/targets/aarch64-linux/lib/libcudart.so.10.2
卸载老的tensorflow，使用命令：
sudo pip3 uninstall -y tensorflow如果安装正确版本的tensorflow后，输出如下：
/local/jetbot$ python3
Python 3.6.9 (default, Jul 17 2020, 12:50:27)
[GCC 8.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow
2020-07-31 14:13:51.745779: I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.2
>>> quit()
5 安装pytorch
Pytorch在X86平台提供了很好的安装包，所以很容易安装。但是，在Nano上，需要使用Nvidia提供的安装包,参照下面的链接下载:
https://forums.developer.nvidia.com/t/pytorch-for-jetson-nano-version-1-5-0-now-available/72048
下载包:
会出现以下错误:
wget https://nvidia.app.box.com/public/static/9eptse6jyly1ggt9axbja2yrmj6pbarc.whl -O torch-1.6.0-cp36-cp36m-linux_aarch64.whl
--2020-08-14 16:14:07-- https://nvidia.app.box.com/public/static/9eptse6jyly1ggt9axbja2yrmj6pbarc.whl
Resolving nvidia.app.box.com (nvidia.app.box.com)... 203.98.7.65, 2001::c765:853e
Connecting to nvidia.app.box.com (nvidia.app.box.com)|203.98.7.65|:443... failed: Connection timed out.
Connecting to nvidia.app.box.com (nvidia.app.box.com)|2001::c765:853e|:443... failed: Network is unreachable
根本无法访问nvidia.app.box.com。
这时，我们需要从源码自己在Nano上编译pytorch。
5.1源码编译pytorch
步骤参考https://forums.developer.nvidia.com/t/pytorch-for-jetson-nano-version-1-6-0-now-available/72048
5.1.1 下载pytorch源码
运行：$ git clone --recursive --branch v1.6.0 https://github.com.cnpmjs.org/pytorch/pytorch
如果报如下错误：
robin@robin-nano:/local/jetbot$
git clone --recursive --branch v1.6.0 https://github.com.cnpmjs.org/pytorch/pytorch
Cloning into 'pytorch'...
remote: Enumerating objects: 419310, done.
Receiving objects: 11% (47614/419310), 15.38 MiB | 22.00 KiB/s
error: RPC failed; curl 56 GnuTLS recv error (-9): A TLS packet with unexpected length was received.
fatal: The remote end hung up unexpectedly
fatal: early EOF
fatal: index-pack failed
则错误解决方法为运行如下三条命令：
sudo apt-get install gnutls-bin
git config --global http.sslVerify false
git config --global http.postBuffer 1048576000
5.1.2 设置编译参数
$ export USE_NCCL=0$ export USE_DISTRIBUTED=0 # skip setting this if you want to enable OpenMPI backend$ export USE_QNNPACK=0$ export USE_PYTORCH_QNNPACK=0$ exportTORCH_CUDA_ARCH_LIST="5.3;6.2;7.2"$ export PYTORCH_BUILD_VERSION=1.6.0 #without the leading 'v', e.g. 1.3.0 for PyTorch v1.3.05.1.3 编译
5.1.3.1 安装依赖
$ sudo apt-get install python3-pip cmake libopenblas-dev
$ pip3 install -r requirements.txt
如果报如下错误:
Running setup.py bdist_wheel for numpy ... error Complete output from command /usr/bin/python3 -u -c "import setuptools, tokenize;__file__='/tmp/pip-build-s5px6e7p/numpy/setup.py';f=getattr(tokenize, 'open', open)(__file__);code=f.read().replace('\r\n', '\n');f.close();exec(compile(code, __file__, 'exec'))" bdist_wheel -d /tmp/tmpxam0ar65pip-wheel- --python-tag cp36: Running from numpy source directory.Cythonizing sourcesProcessing numpy/random/_bounded_integers.pxd.inProcessing numpy/random/_philox.pyxTraceback (most recent call last):File "/tmp/pip-build-s5px6e7p/numpy/tools/cythonize.py", line 59, in process_pyxfrom Cython.Compiler.Version import version as cython_versionModuleNotFoundError: No module named 'Cython'报此错误，可以运行： $ pip3 install cython然后，重新执行$ pip3 install -r requirements.txt
5.1.3.2安装ninja
$ pip3 install ninja
5.1.3.3编译
$ python3 setup.py bdist_wheel
编译时间非常漫长，我用了差不多12个小时才在nano上编译完成。
编译完成后，安装包位于pytorch/dist/torch-1.6.0a0+b31f58d-cp36-cp36m-linux_aarch64.whl
5.2 安装pytorch
-
sudo pip3 install torch-1.6.0a0+b31f58d-cp36-cp36m-linux_aarch64.whl
-
sudo pip3 install torchvision
5.3 验证安装是否成功
Run $ pip3
>>> import torch
>>> print(torch.__version__)
>>> print('CUDA available: ' + str(torch.cuda.is_available()))
>>> print('cuDNN version: ' + str(torch.backends.cudnn.version()))
>>> a = torch.cuda.FloatTensor(2).zero_()
>>> print('Tensor a = ' + str(a))
>>> b = torch.randn(2).cuda()
>>> print('Tensor b = ' + str(b))
>>> c = a + b
>>> print('Tensor c = ' + str(c))
>>> import torchvision
>>> print(torchvision.__version__)
6 安装traitlets
$sudo python3 -m pip install git+https://github.com/ipython/traitlets@master
7 安装jupyter lab
1) install nodejs and nmp
download from https://nodejs.org/en/download/
install from https://github.com/nodejs/help/wiki/Installation
2) $sudo pip3 install jupyter jupyterlab
3)$sudo jupyter labextension install @jupyter-widgets/jupyterlab-manager
4) $sudo jupyter labextension install @jupyterlab/statusbar
5) $jupyter lab --generate-config
6)$ jupyter notebook password
密码使用：jetbot， 也可以使用别的密码，记住即可， 后面会使用此密码登录jupyter
注意： 系统自带的nodjs版本太低，执行第3）步时会报如下错误，所以要遵循第1）步的方法安装最新版的nodejs
robin@robin-nano:/local/jetbot$ sudo jupyter labextension install @jupyter-widgets/jupyterlab-manager
[sudo] password for robin:
An error occured.
ValueError: Please install nodejs >=10.0.0 before continuing. nodejs may be installed using conda or directly from the nodejs website.
See the log file for details: /tmp/jupyterlab-debug-l4c2378v.log
robin@robin-nano:/local/jetbot$ nodejs -v
v8.10.0
8 安装jebbot repo
1) $sudo apt install python3-smbus2) $git clone https://github.com/NVIDIA-AI-IOT/jetbot3) $cd jetbot $sudo apt-get install cmake4) $sudo python3 setup.py install
9 安装jetbot services
1)$ cd jetbot/utils$python3 create_stats_service.py2) $sudo mv jetbot_stats.service /etc/systemd/system/jetbot_stats.service3) $sudo systemctl enable jetbot_stats4) $sudo systemctl start jetbot_stats查看jetbot_stats是否正常启动：
$ sudo systemctl status jetbot_stats● jetbot_stats.service - JetBot stats display service Loaded: loaded (/etc/systemd/system/jetbot_stats.service; enabled; vendor preset: enabled) Active: active (running) since Fri 2020-08-14 14:36:52 CST; 4s ago Main PID: 11508 (sh) Tasks: 2 (limit: 4174) CGroup: /system.slice/jetbot_stats.service ├─11508 /bin/sh -c python3 -m jetbot.apps.stats └─11515 python3 -m jetbot.apps.stats
5) $python3 create_jupyter_service.py6) $sudo mv jetbot_jupyter.service /etc/systemd/system/jetbot_jupyter.service7) $sudo systemctl enable jetbot_jupyter8) $sudo systemctl start jetbot_jupyter
查看jetbot_jupyter是否正常启动：
$ sudo systemctl status jetbot_jupyter● jetbot_jupyter.service - Jupyter Notebook Service Loaded: loaded (/etc/systemd/system/jetbot_jupyter.service; enabled; vendor preset: enabled) Active: active (running) since Fri 2020-08-14 14:40:01 CST; 5s ago Main PID: 15596 (sh) Tasks: 2 (limit: 4174) CGroup: /system.slice/jetbot_jupyter.service ├─15596 /bin/sh -c jupyter lab --ip=0.0.0.0 --no-browser └─15608 /usr/bin/python3 /usr/local/bin/jupyter-lab --ip=0.0.0.0 --no-browser 8月 14 14:40:01 robin-nano systemd[1]: Started Jupyter Notebook Service.8月 14 14:40:04 robin-nano sh[15596]: [I 14:40:04.619 LabApp] Writing notebook server cookie secret to /home/robin/.local/share/jupyter/runtime/no8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.574 LabApp] JupyterLab extension loaded from /usr/local/lib/python3.6/dist-packages/jupyterlab8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.575 LabApp] JupyterLab application directory is /usr/local/share/jupyter/lab8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.586 LabApp] Serving notebooks from local directory: /home/robin8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.586 LabApp] The Jupyter Notebook is running at:8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.587 LabApp] http://robin-nano:8888/8月 14 14:40:06 robin-nano sh[15596]: [I 14:40:06.587 LabApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmat
10 设置 swapfile
1) $sudo fallocate -l 4G /var/swapfile2) $sudo chmod 600 /var/swapfile3) $sudo mkswap /var/swapfile4) $sudo swapon /var/swapfile5) $sudo bash -c 'echo "/var/swapfile swap swap defaults 0 0" >> /etc/fstab'
11 复制 JetBot notebooks 到home目录
$cp -r ~/jetbot/notebooks ~/Notebooks至此配置完成，重启之后，led屏上会显示当前的IP，磁盘，内存，电量信息。
12使用
按照jebot提供的例子就可以开始学习了。
https://github.com/NVIDIA-AI-IOT/jetbot/wiki/Examples
将nano接入wifi网络，查看IP地址，我这里的IP为192.168.10.137
在另一台机器上打开浏览器，输入192.168.10.137:8888,会进入如下页面，
输入密码后，就可以开始学习了。密码是在第7章时设置的密码。
评论（0）
您还未登录，请登录后发表或查看评论