【mmdeploy】mmdeploy安装（linux + windows）

StevenGerrad

已于 2023-04-24 22:09:21 修改

阅读量2.5k

点赞数 1

分类专栏：深度学习文章标签： windows ubuntu mmdeploy 模型部署深度学习

于 2023-04-24 22:02:26 首次发布

本文链接：https://blog.csdn.net/qq_40690815/article/details/130353261

版权

深度学习专栏收录该内容

11 篇文章 0 订阅

订阅专栏

目前文中还有ubuntu下最后编译Demo的bug未解决。

安装mmdeploy（v0.6.0）——linux

本文参考官方文档《Linux-x86_64 下构建方式》，注意：本文主要采用的是这种构建方式，辅之以《get_start》的构建方式，而后者提到官方提供了预编译包，更加方便一些，但本文在搞了一段时间后才发现提供了预编译包，因为并未采用这种方式。

安装构建和编译工具链

查看cmake版本:cmake --version
查看ubuntu版本：uname -a或者cat /etc/*release

原本似乎没有cmake，gcc版本也不够，按照教程进行安装/升级。（注意cmake和后面要下载的tensorRT等库还是找个别的位置装整洁一点）。

wget https://github.com/Kitware/CMake/releases/download/v3.20.0/cmake-3.20.0-linux-x86_64.tar.gz
tar -xzvf cmake-3.20.0-linux-x86_64.tar.gz
sudo ln -sf $(pwd)/cmake-3.20.0-linux-x86_64/bin/* /usr/bin/

sudo apt-get update
sudo apt-get install gcc-7
sudo apt-get install g++-7

安装依赖包

conda create -n mmdeploy python=3.7 -y
conda activate mmdeploy

conda install pytorch==1.8.0 torchvision==0.9.0 cudatoolkit=11.1 -c pytorch -c conda-forge

export cu_version=cu111 # cuda 11.1
export torch_version=torch1.8
pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/${cu_version}/${torch_version}/index.html

安装 MMDeploy SDK 依赖

OpenCV

安装：

sudo apt-get install libopencv-dev

报错:

E: Unable to correct problems, you have held broken packages.

为了解决这个问题，找了个博客，先进行sudo apt update，发现有报错:

(mmdeploy) root@2bc997559bdb:/home/wangjy/research/mmdeploy# sudo apt update
Hit:1 http://mirrors.aliyun.com/ubuntu bionic InRelease
Ign:3 https://developer.download.nvidia.cn/compute/machine-learning/repos/ubuntu2004/x86_64  InRelease
Get:2 https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2004/x86_64  InRelease [1581 B]
Hit:4 https://developer.download.nvidia.cn/compute/machine-learning/repos/ubuntu2004/x86_64  Release
Err:2 https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2004/x86_64  InRelease
  The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC
Reading package lists... Done
W: GPG error: https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2004/x86_64  InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC
E: The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64  InRelease' is not signed.
N: Updating from such a repository can't be done securely, and is therefore disabled by default.
N: See apt-secure(8) manpage for repository creation and user configuration details.

注意其中倒数第三行的E: ...，我通过博客《
Notice: CUDA Linux Repository Key Rotation》和《apt-update in Azure Nvidia gives publickey error》解决了这一问题，具体代码（在第二个stackoverflow的博客中）：

apt-key del 7fa2af80
rm /etc/apt/sources.list.d/cuda.list
rm /etc/apt/sources.list.d/nvidia-ml.list
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-keyring_1.0-1_all.deb
dpkg -i cuda-keyring_1.0-1_all.deb

但是这一块不解决实际问题，还是有上面的held broken packages报错，搞了一天，最后百度出来一些博客是因为要换正确版本（我的版本是Ubuntu20.04，对应Codename为focal）的安装源。

pplcv

git clone https://github.com/openppl-public/ppl.cv.git
cd ppl.cv
export PPLCV_DIR=$(pwd)
git checkout tags/v0.6.2 -b v0.6.2
./build.sh cuda

本来最后一步报错，但搁置了一段时间后，可能是解决了libopencv-dev和pip install pycuda的问题后它也解决了。

安装推理引擎

ONNXRuntime

pip install onnxruntime==1.8.1

# 注意安装的目录，不要跟项目代码放在一起
wget https://github.com/microsoft/onnxruntime/releases/download/v1.8.1/onnxruntime-linux-x64-1.8.1.tgz
tar -zxvf onnxruntime-linux-x64-1.8.1.tgz
cd onnxruntime-linux-x64-1.8.1
export ONNXRUNTIME_DIR=$(pwd)
export LD_LIBRARY_PATH=$ONNXRUNTIME_DIR/lib:$LD_LIBRARY_PATH

TensorRT（+ cuDNN）

版本刚好和他的教程对得上，但要先去英伟达官网下载对应文件：

tar -zxvf TensorRT-8.2.3.0.Linux.x86_64-gnu.cuda-11.4.cudnn8.2.tar.gz
pip install TensorRT-8.2.3.0/python/tensorrt-8.2.3.0-cp37-none-linux_x86_64.whl
export TENSORRT_DIR=$(pwd)/TensorRT-8.2.3.0
export LD_LIBRARY_PATH=$TENSORRT_DIR/lib:$LD_LIBRARY_PATH
pip install pycuda

cd /the/path/of/cudnn/tgz/file
tar -zxvf cudnn-11.3-linux-x64-v8.2.1.32.tgz
export CUDNN_DIR=$(pwd)/cuda
export LD_LIBRARY_PATH=$CUDNN_DIR/lib64:$LD_LIBRARY_PATH

刚才的pip install pycuda会报错：

ERROR: Could not build wheels for pycuda, which is required to install pyproject.toml-based projects

有的博客说是没有制定pycuda的版本，因而与cuda等的版本对不上，但我看下载的.whl文件的网址上居然没有cuda11.1的对应版本，暂时没找到解决方案，也就是说TensorRT差pycuda没装完，这个码在这。

发现之前的gcc版本没用上，使用gcc --version查看还是gcc-6的版本，不是gcc-7，折腾了一阵后使用which gcc发现这个软连接的指向是定死在gcc-6的，遂进行修改。

(mmdeploy) root@2bc997559bdb:/usr/local/bin# ll
total 12
drwxr-xr-x 1 root root 4096 Jan 17  2021 ./
drwxr-xr-x 1 root root 4096 Jan 19  2021 ../
lrwxrwxrwx 1 root root   14 Jan 17  2021 g++ -> /usr/bin/g++-6*
lrwxrwxrwx 1 root root   14 Jan 17  2021 gcc -> /usr/bin/gcc-6*
(mmdeploy) root@2bc997559bdb:/usr/local/bin# ln -fs /usr/bin/gcc-7* gcc
(mmdeploy) root@2bc997559bdb:/usr/local/bin# ll
total 12
drwxr-xr-x 1 root root 4096 Jul 21 13:03 ./
drwxr-xr-x 1 root root 4096 Jan 19  2021 ../
lrwxrwxrwx 1 root root   14 Jan 17  2021 g++ -> /usr/bin/g++-6*
lrwxrwxrwx 1 root root   14 Jul 21 13:03 gcc -> /usr/bin/gcc-7*
(mmdeploy) root@2bc997559bdb:/usr/local/bin# ln -fs /usr/bin/g++-7* g++
(mmdeploy) root@2bc997559bdb:/usr/local/bin#

但是这个并没有解决pycuda的安装问题。后续又做了一些工作，才成功安装。包括之前的cuda没有nvcc，重新安装了cuda ToolKit，安装libevent-dev，修改~/.bashrc文件以及用source ~/.bashrc重新激活该文件等操作。具体啥起了作用不知道了，甚至可能是之前换ubuntu的安装源起的作用。

后续

需要将上述的临时环境变量添加到~/.bashrc中，以onnx为例：

echo '# set env for onnxruntime' >> ~/.bashrc
echo "export ONNXRUNTIME_DIR=${ONNXRUNTIME_DIR}" >> ~/.bashrc
echo "export LD_LIBRARY_PATH=$ONNXRUNTIME_DIR/lib:$LD_LIBRARY_PATH" >> ~/.bashrc
source ~/.bashrc

编译mmdeploy

cd /the/root/path/of/MMDeploy
export MMDEPLOY_DIR=$(pwd)

ONNXRuntime 自定义算子

cd ${MMDEPLOY_DIR}
mkdir -p build && cd build
cmake -DCMAKE_CXX_COMPILER=g++-7 -DMMDEPLOY_TARGET_BACKENDS=ort -DONNXRUNTIME_DIR=${ONNXRUNTIME_DIR} ..
make -j$(nproc) && make install

其中有一些报错如：

CMake Error at csrc/mmdeploy/core/CMakeLists.txt:15 (add_subdirectory):
  The source directory

    /home/wangjy/research/mmdeploy/third_party/spdlog

  does not contain a CMakeLists.txt file.

找来找去找到了一个命令git submodule update --init --recursive，并且在一个issue——mmdeploy/issues/260里确认了需要这个命令

安装 Model Converter

cd ${MMDEPLOY_DIR}
pip install -e .

编译SDK（需要安装 MMDeploy SDK 依赖）

cpu + ONNXRuntime

cd ${MMDEPLOY_DIR}
mkdir -p build && cd build
cmake .. \
    -DCMAKE_CXX_COMPILER=g++-7 \
    -DMMDEPLOY_BUILD_SDK=ON \
    -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON \
    -DMMDEPLOY_TARGET_DEVICES=cpu \
    -DMMDEPLOY_TARGET_BACKENDS=ort \
    -DMMDEPLOY_CODEBASES=all \
    -DONNXRUNTIME_DIR=${ONNXRUNTIME_DIR}

make -j$(nproc) && make install

cuda + TensorRT

cd ${MMDEPLOY_DIR}
mkdir -p build && cd build
cmake .. \
    -DCMAKE_CXX_COMPILER=g++-7 \
    -DMMDEPLOY_BUILD_SDK=ON \
    -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON \
    -DMMDEPLOY_TARGET_DEVICES="cuda;cpu" \
    -DMMDEPLOY_TARGET_BACKENDS=trt \
    -DMMDEPLOY_CODEBASES=all \
    -Dpplcv_DIR=${PPLCV_DIR}/cuda-build/install/lib/cmake/ppl \
    -DTENSORRT_DIR=${TENSORRT_DIR} \
    -DCUDNN_DIR=${CUDNN_DIR}

make -j$(nproc) && make install

编译 Demo

cd ${MMDEPLOY_DIR}/build/install/example
mkdir -p build && cd build
cmake .. -DMMDeploy_DIR=${MMDEPLOY_DIR}/build/install/lib/cmake/MMDeploy
make -j$(nproc)

最后一个命令报错了：

/usr/bin/ld: /usr/lib/gcc/x86_64-linux-gnu/7/libstdc++fs.a(ops.o): undefined reference to symbol '_ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE6resizeEmc'
/usr/bin/ld: /home/wangjy/research/mmdeploy/build/install/lib/libmmdeploy_core.so: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status

网上有说换CMake_Cache.txt的CXX编译源换成g++的，没用；有说使用c++filt看一下报错对应语句，不过不知道有啥用，并没有解决问题：

(mmdeploy) root@2bc997559bdb:/# c++filt _ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE6resizeEmc
std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >::resize(unsigned long, char)

编译mmdeploy——使用g++ -9

cd ${MMDEPLOY_DIR}
mkdir -p build && cd build

cmake -DCMAKE_CXX_COMPILER=g++-9 -DMMDEPLOY_TARGET_BACKENDS=ort -DONNXRUNTIME_DIR=${ONNXRUNTIME_DIR} ..
make -j$(nproc) && make install

cmake -DCMAKE_CXX_COMPILER=g++-9 -DMMDEPLOY_TARGET_BACKENDS=trt -DTENSORRT_DIR=${TENSORRT_DIR} -DCUDNN_DIR=${CUDNN_DIR} ..
make -j$(nproc) && make install

cd ${MMDEPLOY_DIR}
pip install -e .

cd build
cmake .. \
    -DCMAKE_CXX_COMPILER=g++-9 \
    -DMMDEPLOY_BUILD_SDK=ON \
    -DMMDEPLOY_BUILD_EXAMPLES=ON \
    -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON \
    -DMMDEPLOY_TARGET_DEVICES=cpu \
    -DMMDEPLOY_TARGET_BACKENDS=ort \
    -DMMDEPLOY_CODEBASES=all \
    -DONNXRUNTIME_DIR=${ONNXRUNTIME_DIR}
make -j$(nproc) && make install

cmake .. \
    -DCMAKE_CXX_COMPILER=g++-9 \
    -DMMDEPLOY_BUILD_SDK=ON \
    -DMMDEPLOY_BUILD_EXAMPLES=ON \
    -DMMDEPLOY_BUILD_SDK_PYTHON_API=ON \
    -DMMDEPLOY_TARGET_DEVICES="cuda;cpu" \
    -DMMDEPLOY_TARGET_BACKENDS=trt \
    -DMMDEPLOY_CODEBASES=all \
    -Dpplcv_DIR=${PPLCV_DIR}/cuda-build/install/lib/cmake/ppl \
    -DTENSORRT_DIR=${TENSORRT_DIR} \
    -DCUDNN_DIR=${CUDNN_DIR}
make -j$(nproc) && make install

安装mmdeploy(v0.6.0)——Windows

不知道为什么windows安装文档开头没有说安装python。而后面安装tensorRT的时候会发现python版本最好为3.7.0，而那个时候再安装python==3.7.0要转半天。

conda create -n mmdeploy python=3.7 -y
conda activate mmdeploy

PS: 注意后面的一些使用$env:添加环境变量的方式只能留在对应的命令行窗口中，这里整理了需要添加的环境变量。

$env:OpenCV_DIR="E:\ProgramData\OpenCV\opencv\build"
$env:path = "\$env:OpenCV_DIR\x64\vc15\bin" + "$env:path"
$env:PPLCV_DIR = "E:\ProgramData\ppl.cv"
$env:ONNXRUNTIME_DIR = "E:\ProgramData\onnxruntime-win-x64-1.8.1"
$env:path = "$env:ONNXRUNTIME_DIR\lib;" + $env:path
$env:TENSORRT_DIR = "E:\ProgramData\TensorRT-8.2.3.0"
$env:path = "$env:TENSORRT_DIR\lib;" + $env:path
$env:CUDNN_DIR="E:\ProgramData\cuda"
$env:path = "$env:CUDNN_DIR\bin;" + $env:path
$env:MMDEPLOY_DIR="E:\project\mmdeploy"
$env:path = "$env:MMDEPLOY_DIR/build/install/bin;" + $env:path

安装构建和编译工具链

CUDA安装包

安装依赖包

安装 MMDeploy Converter 依赖

conda 略
PyTorch + mmcv-full

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge

pip install mmcv-full==1.4.0 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.10.0/index.html

安装 MMDeploy SDK 依赖

OpenCV
在给出的网址找了一个.exe文件安装后将OpenCVConfig.cmake所在路径添加到环境变量PATH中(参考博客)。

# 在电脑->属性->环境变量->系统变量->PATH中添加
E:\ProgramData\OpenCV\opencv\build
E:\ProgramData\OpenCV\opencv\build\x64\vc15\bin

$env:path = "E:\ProgramData\OpenCV\opencv\build" + "$env:path"
$env:path = "E:\ProgramData\OpenCV\opencv\build\x64\vc15\bin" + "$env:path"

$env:OpenCV_DIR="E:\ProgramData\OpenCV\opencv\build"

pplcv

build v0.6.2 or v0.6.3时会有报错：

-- Configuring done
CMake Error at CMakeLists.txt:48 (add_library):
  No SOURCES given to target: pplcv_static

CMake Generate step failed.  Build files cannot be regenerated correctly.

而v0.7.0没有报错，要问为什么要去build前两者，因为最后build tensorRT时有报错与前两者有关…

PS: 查看环境变量：ls env:

安装推理引擎

略

编译 MMDeploy

编译安装 Model Converter

编译自定义算子
安装 Model Converter

PS：适当的时候可能还是要和ubuntu一样进行

cd ${MMDEPLOY_DIR}
git checkout tags/v0.6.0
git submodule update --init --recursive

编译 SDK

OpenCV相关报错：

-- CMAKE_INSTALL_PREFIX: E:/project/mmdeploy/build/install
-- Selecting Windows SDK version 10.0.19041.0 to target Windows 10.0.18363.
-- Build ONNXRUNTIME custom ops.
CMake Error at cmake/opencv.cmake:2 (find_package):
  By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "OpenCV", but
  CMake did not find one.

  Could not find a package configuration file provided by "OpenCV" with any
  of the following names:

    OpenCVConfig.cmake
    opencv-config.cmake

  Add the installation prefix of "OpenCV" to CMAKE_PREFIX_PATH or set
  "OpenCV_DIR" to a directory containing one of the above files.  If "OpenCV"
  provides a separate development package or SDK, be sure it has been
  installed.
Call Stack (most recent call first):
  csrc/mmdeploy/CMakeLists.txt:7 (include)


-- Configuring incomplete, errors occurred!
See also "E:/project/mmdeploy/build/CMakeFiles/CMakeOutput.log".

最后似乎是在上文添加环境变量解决的$env:OpenCV_DIR="E:\ProgramData\OpenCV\opencv\build"。

编译SDK——cuda + TensorRT时，pplcv相关报错：

By not providing "Findpplcv.cmake" in CMAKE_MODULE_PATH this project has asked CMake to find a package configuration file provided by "pplcv", but CMake did not find one.

Could not find a package configuration file provided by "pplcv" with any of the following names:
pplcvConfig.cmake
pplcv-config.cmake

重新安装pplcv，注意观察编译的cmake语句：

cmake .. -G "Visual Studio 16 2019" -A x64 -T v142 `
  -DMMDEPLOY_BUILD_SDK=ON `
  -DMMDEPLOY_TARGET_DEVICES="cuda" `
  -DMMDEPLOY_TARGET_BACKENDS="trt" `
  -DMMDEPLOY_CODEBASES="all" `
  -Dpplcv_DIR="$env:PPLCV_DIR/pplcv-build/install/lib/cmake/ppl" `
  -DTENSORRT_DIR="$env:TENSORRT_DIR" `
  -DCUDNN_DIR="$env:CUDNN_DIR"

即注意之前安装的ppl.cv项目中一定要有pplcv-build/install/lib/cmake/ppl没有他就会报上面的错误，或者可能是其他pplcv相关的error，说明安装的不完整。

但是还是有pplcv相关的报错：

E:\project\mmdeploy\csrc\mmdeploy\preprocess\cuda\pad_impl.cpp(24): error C2039: "BORDER_TYPE_CONSTANT": 不是 "ppl::cv" 的成员 [E:\proje
ct\mmdeploy\build\csrc\mmdeploy\preprocess\cuda\mmdeploy_cuda_transform_impl.vcxproj]
  E:\ProgramData\ppl.cv\pplcv-build\install\lib\cmake\ppl\..\..\..\include\ppl/cv/cuda/copymakeborder.h(26): note: 参见“ppl::cv”的声明
E:\project\mmdeploy\csrc\mmdeploy\preprocess\cuda\pad_impl.cpp(24): error C2065: “BORDER_TYPE_CONSTANT”: 未声明的标识符 [E:\project\mmdepl
oy\build\csrc\mmdeploy\preprocess\cuda\mmdeploy_cuda_transform_impl.vcxproj]

参考issues/824解决该问题，即将E:\ProgramData\ppl.cv\pplcv-build\install\lib\cmake\ppl中的对应项改为：

set(PPLCV_VERSION_MAJOR 0)
set(PPLCV_VERSION_MINOR 6)
set(PPLCV_VERSION_PATCH 2)

但是在使用下文的编译demo时有错误：

[2022-07-31 11:40:37.623] [mmdeploy] [error] [net_module.cpp:37] Net backend not found: tensorrt
[2022-07-31 11:40:37.623] [mmdeploy] [error] [task.cpp:67] error parsing config: {
  "context": {
    "device": "<any>",
    "model": "<any>",
    "stream": "<any>"
  },
  "input": [
    "prep_output"
  ],
  "input_map": {
    "img": "input"
  },
  "module": "Net",
  "name": "maskrcnn",
  "output": [
    "infer_output"
  ],
  "type": "Task"
}

cmake .. -G "Visual Studio 16 2019" -A x64 -T v142 `
  -DMMDEPLOY_BUILD_SDK=ON `
  -DMMDEPLOY_TARGET_DEVICES="cuda" `
  -DMMDEPLOY_TARGET_BACKENDS="ort;trt" `
  -DMMDEPLOY_CODEBASES="all" `
  -Dpplcv_DIR="$env:PPLCV_DIR/pplcv-build/install/lib/cmake/ppl" `
  -DTENSORRT_DIR="$env:TENSORRT_DIR" `
  -DCUDNN_DIR="$env:CUDNN_DIR"