JetPack SDK Packages安装（非SDKManager）

如下教程可分步骤单独安装SDK Packages，此教程对应的JetPack版本是JetPack4.4-L4T-R32.4.3/JetPack4.4.1-L4T-R32.4.4（如需要其他版本请参考此方法，下载对应版本即可）
JetPack4.4 SDK Packages：(百度网盘共享链接）：
链接： https://pan.baidu.com/s/1XqLujhw5dc423WBGJsXcgg
提取码：x7ck

0. 查看jetson 版本（命令：jetson_release）

安装jtop 工具

nvidia@nx:~$ jetson_release 
 - NVIDIA Jetson Xavier NX
   * Jetpack 4.4 [L4T 32.4.3]
   * NV Power Mode: MODE_10W_2CORE - Type: 3
   * jetson_clocks service: inactive
 - Libraries:
   * CUDA: 10.2.89
   * cuDNN: 8.0.0.180
   * TensorRT: NOT_INSTALLED
   * Visionworks: NOT_INSTALLED
   * OpenCV: NOT_INSTALLED compiled CUDA: NO
   * VPI: NOT_INSTALLED
   * Vulkan: 1.2.70
nvidia@nx:~$ 

1. CUDA-10.2安装 JetPack4.4/4.4.1 CUDA版本相同

$sudo dpkg -i /opt/nvidia/deb_repos/cuda-repo-l4t-10-2-local-10.2.89_1.0-1_arm64.deb 

#安装完成会提示pub key， 根据提示添加apt key(.pub)，例如：
$sudo apt-key add /var/cuda-repo-10-2-local-10.2.89/7fa2af80.pub
$sudo apt-get -y update 
$sudo apt-get -y  install cuda-toolkit-10-2 

#以上安装完成后，通过nvcc 查询不到，但可以搜索目录/usr/local 是否有cuda，此时可通过添加环境变量 ~/.bash.rc 
$vi ~/.bashrc 
export PATH=/usr/local/cuda-10.2/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:$LD_LIBRARY_PATH

nvidia@nx:~$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_21:14:42_PDT_2019
Cuda compilation tools, release 10.2, V10.2.89

2. cuDNN-8.0安装 JetPack4.4/4.4.1 cuDNN版本相同

$sudo dpkg -i libcudnn8_8.0.0.180-1+cuda10.2_arm64.deb
$sudo dpkg -i libcudnn8-dev_8.0.0.180-1+cuda10.2_arm64.deb
$sudo dpkg -i libcudnn8-doc_8.0.0.180-1+cuda10.2_arm64.deb
$sudo apt-get -y update 
以上安装完成后，通过jetson_release 查看安装后的版本信息
$ jetson_release 

安装CUDA和cuDNN 后， 剩余空间约2.8GB

3. TensorRT-7.13安装 JetPack4.4/4.4.1 TensorRT版本相同

$sudo dpkg -i libnvinfer7_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvinfer-plugin7_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvinfer-plugin-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvonnxparsers7_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvonnxparsers-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvparsers7_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvparsers-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvinfer-bin_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i libnvinfer-doc_7.1.3-1+cuda10.2_all.deb 
$sudo dpkg -i libnvinfer-samples_7.1.3-1+cuda10.2_all.deb 
$sudo dpkg -i tensorrt_7.1.3.0-1+cuda10.2_arm64.deb 
$sudo dpkg -i python-libnvinfer_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i python-libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i python3-libnvinfer_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i python3-libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i graphsurgeon-tf_7.1.3-1+cuda10.2_arm64.deb 
$sudo dpkg -i uff-converter-tf_7.1.3-1+cuda10.2_arm64.deb 
或
$sudo dpkg -i libnvinfer7_7.1.3-1+cuda10.2_arm64.deb libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb libnvinfer-plugin7_7.1.3-1+cuda10.2_arm64.deb libnvinfer-plugin-dev_7.1.3-1+cuda10.2_arm64.deb libnvonnxparsers7_7.1.3-1+cuda10.2_arm64.deb libnvonnxparsers-dev_7.1.3-1+cuda10.2_arm64.deb libnvparsers7_7.1.3-1+cuda10.2_arm64.deb libnvparsers-dev_7.1.3-1+cuda10.2_arm64.deb libnvinfer-bin_7.1.3-1+cuda10.2_arm64.deb libnvinfer-doc_7.1.3-1+cuda10.2_all.deb libnvinfer-samples_7.1.3-1+cuda10.2_all.deb tensorrt_7.1.3.0-1+cuda10.2_arm64.deb python-libnvinfer_7.1.3-1+cuda10.2_arm64.deb python-libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb python3-libnvinfer_7.1.3-1+cuda10.2_arm64.deb python3-libnvinfer-dev_7.1.3-1+cuda10.2_arm64.deb graphsurgeon-tf_7.1.3-1+cuda10.2_arm64.deb uff-converter-tf_7.1.3-1+cuda10.2_arm64.deb 

$sudo apt-get -y update 
以上安装完成后，通过jetson_release 查看安装后的版本信息
$ jetson_release 

4. OpenCV 安装

由于NVIDIA SDKManager自带安装的OpenCV都不支持CUDA， 所以安装意义不大，不如通过源码编译安装，如下介绍NX/Nano的OpenCV 源码编译安装！
参考：Nano_build_opencv

$git clone https://github.com/mdegans/nano_build_opencv.git
$cd nano_build_opencv/
$./build_opencv.sh 4.3.0   
#指定版本编译，下载时间一般较长，把如下一句指令注释掉可节约一些时间
#sudo apt-get dist-upgrade -y --autoremove

问题1、 编译opencv_contrib或opencv时提示缺少boostdesc_bgm.i等编译错误
错误：
opencv_contrib/modules/xfeatures2d/src/boostdesc.cpp:673:20: fatal error: boostdesc_bgm.i: No such file or directory
解决：
由于采用的是opencv源码编译方式，可查看 build文件夹下的日志文件 CMakeDownloadLog.txt并搜索 boostdesc_bgm.i 关键词 ，发现这个文件下载失败同时还有其他一些.i 文件下载识别，此txt日志文件中有它们的下载地址，直接复制其下载地址到网页可以看该到文件的源码，可直接拷贝源码并保存同名文件，存放于opencv_contrib/modules/xfeatures2d/src/ 路径下，文件包含：
boostdesc_bgm.i
boostdesc_bgm_bi.i
boostdesc_bgm_hd.i
boostdesc_lbgm.i
boostdesc_binboost_064.i
boostdesc_binboost_128.i
boostdesc_binboost_256.i
vgg_generated_120.i
vgg_generated_64.i
vgg_generated_80.i
vgg_generated_48.i

或 通过上面百度网盘分享的opencv 4.3.0 源码编译安装

下载并解压缩：opencv.gz、opencv_contrib.gz、build_opencv.sh 
$mkdir -p /tmp/build_opencv
$cd /tmp/build_opencv
$tar zxvf opencv.gz
$tar zxvf opencv_contrib.gz 
$./build_opencv.sh

附录：
安装完成，第一件事把已安装好的镜像备份出来，备份及升级方法参考连接：
Jetson个平台系统升级命令合集

5. DeepStream 安装

deepstream-getting-started

下载tar 或 deb 安装包进行安装即可
https://developer.nvidia.com/assets/Deepstream/5.0/ga/secure/deepstream_sdk_5.0_jetson.tbz2
https://developer.nvidia.com/assets/Deepstream/5.0/ga/secure/deepstream_sdk_5.0_arm64.deb

DeepStream Apps
https://github.com/NVIDIA-AI-IOT/deepstream_python_apps