jetson-inference在ubuntu的x86_64环境搭建

绯虹剑心

已于 2023-05-08 16:54:29 修改

阅读量428

点赞数

CC 4.0 BY-SA版权

分类专栏：其他记录文章标签： ubuntu

于 2023-05-08 15:32:15 首次发布

本文链接：https://blog.youkuaiyun.com/weixin_43422012/article/details/130556422

其他记录专栏收录该内容

8 篇文章

订阅专栏

克隆代码

git clone https://github.com/dusty-nv/jetson-inference
cd jetson-inference
git submodule update --init

如果超时git clone多试几次

防止子模块下载超时
.gitmodules 文件中
https://github.com 换成 https://github.com.cnpmjs.org

git submodule sync
git submodule update --init --recursive

如果下载失败，可以git status 有差异的是失败目录，需要手动删除再执行更新命令

cmake

cd jetson-inference   
mkdir build
cd build
cmake ../

cmake 会报错，需安装cuda,tensorrt,cudnn

安装cuda

https://developer.nvidia.com/cuda-toolkit-archive
linux x86_64 ubuntu 20.04 runfile

在/etc/profile 结尾加环境变量

export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib:/usr/local/cuda/lib64

source /etc/profile
ldconfig

vim ~/.bashrc

在.bashrc结尾添加

source /etc/profile

如果有conda环境，在.bashrc结尾添加

conda activate base

用命令查看版本

nvcc -V

安装tensorrt

下载和cuda匹配的版本
https://developer.nvidia.com/nvidia-tensorrt-8x-download

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/lib:/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/targets/x86_64-linux-gnu/lib
export C_INCLUDE_PATH=$C_INCLUDE_PATH:/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/include
export CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/include
export CUDA_MODULE_LOADING=LAZY

修改 CMakeLists.txt
在 # build C/C++ library行后引入TensorRT依赖

include_directories(/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/include)
link_directories(/workspace/doc/nvidia/TensorRT/TensorRT-8.6.0.12/lib/)

安装cudnn

下载和cuda匹配的版本
https://developer.nvidia.com/rdp/cudnn-archive
解压后把文件复制到cuda

cd include
mv *.* /usr/local/cuda/include
cd lib
mv *.* /usr/local/cuda/lib64

make

cd jetson-inference   
cd build
make 
make install

运行示例

cd /build/x86_64/bin
./imagenet.py --network=vgg-16 images/jellyfish.jpg images/test/output_jellyfish.jpg

会报下载模型失败的错误，运行如下命令,修改下载地址和下载模型

cd jetson-inference/tools/
sed -in-place -e 's@https://nvidia.box.com/shared/static@https://bbs.gpuworld.cn/mirror@g' download-models.sh
./download-models.sh

或参考 https://code84.com/763863.html