Mask R-CNN pythoch版windows环境配置

硬件需求：
win10系统，pycharm，NIVIDA显卡以及下载好的Cuda

---

源码地址：https://github.com/facebookresearch/maskrcnn-benchmark

---

前言

打开命令提示符，输入命令nvcc --version，查看显卡配置：

C:\WINDOWS\system32>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:04_Central_Daylight_Time_2018
Cuda compilation tools, release 10.0, V10.0.130

我的cuda安装的是10.0版本，在后续可以进行参考。

---

提示：以下是本篇文章正文内容，下面案例可供参考

一、创建虚拟环境

输入命令，新建一共名为maskrcnn_pytorch的虚拟环境

conda create -n maskrcnn_pytorch python=3.7 pip

出现如下提示，选择y

## Package Plan ##

  environment location: D:\Anaconda\program\envs\maskrcnn_pytorch

  added / updated specs:
    - pip
    - python=3.7


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    certifi-2021.5.30          |   py37haa95532_0         142 KB  https://mirrors.ustc.edu.cn/anaconda/pkgs/main
    pip-21.1.3                 |   py37haa95532_0         2.0 MB  https://mirrors.ustc.edu.cn/anaconda/pkgs/main
    python-3.7.11              |       h6244533_0        17.4 MB  https://mirrors.ustc.edu.cn/anaconda/pkgs/main
    sqlite-3.36.0              |       h2bbff1b_0         1.2 MB  https://mirrors.ustc.edu.cn/anaconda/pkgs/main
    ------------------------------------------------------------
                                           Total:        20.7 MB

The following NEW packages will be INSTALLED:

  ca-certificates    anaconda/pkgs/main/win-64::ca-certificates-2021.7.5-haa95532_1
  certifi            anaconda/pkgs/main/win-64::certifi-2021.5.30-py37haa95532_0
  openssl            anaconda/pkgs/main/win-64::openssl-1.1.1k-h2bbff1b_0
  pip                anaconda/pkgs/main/win-64::pip-21.1.3-py37haa95532_0
  python             anaconda/pkgs/main/win-64::python-3.7.11-h6244533_0
  setuptools         anaconda/pkgs/main/win-64::setuptools-52.0.0-py37haa95532_0
  sqlite             anaconda/pkgs/main/win-64::sqlite-3.36.0-h2bbff1b_0
  vc                 anaconda/pkgs/main/win-64::vc-14.2-h21ff451_1
  vs2015_runtime     anaconda/pkgs/main/win-64::vs2015_runtime-14.27.29016-h5e58377_2
  wheel              anaconda/pkgs/main/noarch::wheel-0.36.2-pyhd3eb1b0_0
  wincertstore       anaconda/pkgs/main/win-64::wincertstore-0.2-py37_0


Proceed ([y]/n)?

如果使用国内镜像的话，大约30s安装完毕，显示

done
#
# To activate this environment, use
#
#     $ conda activate maskrcnn_pytorch
#
# To deactivate an active environment, use
#
#     $ conda deactivate

进入虚拟环境

conda activate maskrcnn_pytorch

出现以下的字样：

(maskrcnn_pytorch) C:\WINDOWS\system32>

虚拟环境创建完毕

二、安装对应库

1.pytorch

首先安装pytorch，pytorch需要与自己的cuda相对应，我的cuda是10.0版本，同时，Mask R-CNN环境要求pytorch在1.0以上，在pytorch官网找到对应的版本
官网地址：
https://pytorch.org/get-started/previous-versions/
我们选择在线安装

# CUDA 9.0
conda install pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=9.0 -c pytorch

# CUDA 10.0
conda install pytorch==1.0.1 torchvision==0.2.2 cudatoolkit=10.0 -c pytorch

# CPU Only
conda install pytorch-cpu==1.0.1 torchvision-cpu==0.2.2 cpuonly -c pytorch

选择对应的命令，出现：

## Package Plan ##

  environment location: D:\Anaconda\program\envs\maskrcnn_pytorch

  added / updated specs:
    - cudatoolkit=10.0
    - pytorch==1.0.1
    - torchvision==0.2.2


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    cudatoolkit-10.0.130       |                0       371.0 MB  https://mirrors.ustc.edu.cn/anaconda/pkgs/main
    pytorch-1.0.1              |py3.7_cuda100_cudnn7_1       464.9 MB  pytorch
    torchvision-0.2.2          |             py_3          44 KB  pytorch
    ------------------------------------------------------------
                                           Total:       835.9 MB

The following NEW packages will be INSTALLED:

  blas               anaconda/pkgs/main/win-64::blas-1.0-mkl
  cffi               anaconda/pkgs/main/win-64::cffi-1.14.6-py37h2bbff1b_0
  cudatoolkit        anaconda/pkgs/main/win-64::cudatoolkit-10.0.130-0
  freetype           anaconda/pkgs/main/win-64::freetype-2.10.4-hd328e21_0
  intel-openmp       anaconda/pkgs/main/win-64::intel-openmp-2021.3.0-haa95532_3372
  jpeg               anaconda/pkgs/main/win-64::jpeg-9b-hb83a4c4_2
  libpng             anaconda/pkgs/main/win-64::libpng-1.6.37-h2a8f88b_0
  libtiff            anaconda/pkgs/main/win-64::libtiff-4.2.0-hd0e1b90_0
  lz4-c              anaconda/pkgs/main/win-64::lz4-c-1.9.3-h2bbff1b_0
  mkl                anaconda/pkgs/main/win-64::mkl-2021.3.0-haa95532_524
  mkl-service        anaconda/pkgs/main/win-64::mkl-service-2.4.0-py37h2bbff1b_0
  mkl_fft            anaconda/pkgs/main/win-64::mkl_fft-1.3.0-py37h277e83a_2
  mkl_random         anaconda/pkgs/main/win-64::mkl_random-1.2.2-py37hf11a4ad_0
  ninja              anaconda/pkgs/main/win-64::ninja-1.10.2-h6d14046_1
  numpy              anaconda/pkgs/main/win-64::numpy-1.20.3-py37ha4e8547_0
  numpy-base         anaconda/pkgs/main/win-64::numpy-base-1.20.3-py37hc2deb75_0
  olefile            anaconda/pkgs/main/win-64::olefile-0.46-py37_0
  pillow             anaconda/pkgs/main/win-64::pillow-8.3.1-py37h4fa10fc_0
  pycparser          anaconda/pkgs/main/noarch::pycparser-2.20-py_2
  pytorch            pytorch/win-64::pytorch-1.0.1-py3.7_cuda100_cudnn7_1
  six                anaconda/pkgs/main/noarch::six-1.16.0-pyhd3eb1b0_0
  tk                 anaconda/pkgs/main/win-64::tk-8.6.10-he774522_0
  torchvision        pytorch/noarch::torchvision-0.2.2-py_3
  xz                 anaconda/pkgs/main/win-64::xz-5.2.5-h62dcd97_0
  zlib               anaconda/pkgs/main/win-64::zlib-1.2.11-h62dcd97_4
  zstd               anaconda/pkgs/main/win-64::zstd-1.4.9-h19a0ad4_0


Proceed ([y]/n)?

选择y，这一过程大约等待5至10分钟，pytorch安装完毕。

2.ipython等其余安装包

安装ipython

conda install ipython pip

选择y

## Package Plan ##

  environment location: D:\Anaconda\program\envs\maskrcnn_pytorch

  added / updated specs:
    - ipython
    - pip


The following NEW packages will be INSTALLED:

  backcall           anaconda/pkgs/main/noarch::backcall-0.2.0-pyhd3eb1b0_0
  colorama           anaconda/pkgs/main/noarch::colorama-0.4.4-pyhd3eb1b0_0
  decorator          anaconda/pkgs/main/noarch::decorator-5.0.9-pyhd3eb1b0_0
  ipython            anaconda/pkgs/main/win-64::ipython-7.22.0-py37hd4e2768_0
  ipython_genutils   anaconda/pkgs/main/noarch::ipython_genutils-0.2.0-pyhd3eb1b0_1
  jedi               anaconda/pkgs/main/win-64::jedi-0.17.0-py37_0
  parso              anaconda/pkgs/main/noarch::parso-0.8.2-pyhd3eb1b0_0
  pickleshare        anaconda/pkgs/main/noarch::pickleshare-0.7.5-pyhd3eb1b0_1003
  prompt-toolkit     anaconda/pkgs/main/noarch::prompt-toolkit-3.0.17-pyh06a4308_0
  pygments           anaconda/pkgs/main/noarch::pygments-2.9.0-pyhd3eb1b0_0
  traitlets          anaconda/pkgs/main/noarch::traitlets-5.0.5-pyhd3eb1b0_0
  wcwidth            anaconda/pkgs/main/noarch::wcwidth-0.2.5-py_0


Proceed ([y]/n)?

根据requestment.txt安装其余的库

pip install ninja yacs cython matplotlib tqdm opencv-python

结果显示

Collecting ninja
  Using cached https://pypi.tuna.tsinghua.edu.cn/packages/2c/f1/30316c8ec18ef9837c1c957379110eef28f9d9d08837bd6643066bdcc329/ninja-1.10.2-py2.py3-none-win_amd64.whl (837 kB)
Collecting yacs
  Using cached https://pypi.tuna.tsinghua.edu.cn/packages/38/4f/fe9a4d472aa867878ce3bb7efb16654c5d63672b86dc0e6e953a67018433/yacs-0.1.8-py3-none-any.whl (14 kB)
Collecting cython
  Using cached https://pypi.tuna.tsinghua.edu.cn/packages/4f/1c/d5b7b91f99044bf84da41059912f2a38ce8db9c9b65bfa50e3cf8afa07c5/Cython-0.29.24-cp37-cp37m-win_amd64.whl (1.6 MB)
Requirement already satisfied: matplotlib in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (3.0.2)
Collecting tqdm
  Using cached https://pypi.tuna.tsinghua.edu.cn/packages/0b/e8/d6f4db0886dbba2fc87b5314f2d5127acdc782e4b51e6f86972a2e45ffd6/tqdm-4.62.0-py2.py3-none-any.whl (76 kB)
Requirement already satisfied: opencv-python in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (4.3.0.36)
Collecting PyYAML
  Using cached https://pypi.tuna.tsinghua.edu.cn/packages/66/7a/9f3a49dea61e98ac8bd3b6b619da4fcb59053232a6e11e9e865dab7c16e0/PyYAML-5.4.1-cp37-cp37m-win_amd64.whl (210 kB)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from matplotlib) (2.4.7)
Requirement already satisfied: kiwisolver>=1.0.1 in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from matplotlib) (1.2.0)
Requirement already satisfied: cycler>=0.10 in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from matplotlib) (0.10.0)
Requirement already satisfied: numpy>=1.10.0 in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from matplotlib) (1.17.0+mkl)
Requirement already satisfied: python-dateutil>=2.1 in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from matplotlib) (2.8.1)
Requirement already satisfied: colorama; platform_system == "Windows" in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from tqdm) (0.4.3)
Requirement already satisfied: six in c:\users\guojinyue\appdata\roaming\python\python37\site-packages (from cycler>=0.10->matplotlib) (1.15.0)
Installing collected packages: ninja, PyYAML, yacs, cython, tqdm
Successfully installed PyYAML-5.4.1 cython-0.29.24 ninja-1.10.2 tqdm-4.62.0 yacs-0.1.8

pip库安装成功。

如果在安装中出现：

WARNING conda.gateways.disk.delete:unlink_or_rename_to_trash(140): Could not remove or rename D:\Anaconda\program\pkgs\jedi-0.17.0-py37_0.tar.bz2.  Please remove this file manually (you may need to reboot to free file handles)
WARNING conda.gateways.disk.delete:unlink_or_rename_to_trash(140): Could not remove or rename D:\Anaconda\program\pkgs\jedi-0.17.0-py37_0\Lib\site-packages\jedi\api\__init__.py.  Please remove this file manually (you may need to reboot to free file handles)

这是anaconda3文件夹权限问题，我的anaconda3文件夹安装在D盘，可能安装在默认C盘路径就不会出现问题。
根据提示，手动删除路径上的对应文件，重新输入指令，即可安装成功

3.检验环境

在虚拟环境中进入python，进行环境验证

>>> import torch
>>> torch.__version__
'1.0.1'

>>> torch.cuda.is_available()
True

>>> a=torch.rand(1,2)
>>> b=a.cuda()
>>> print(a,b)
tensor([[0.5375, 0.1968]]) tensor([[0.5375, 0.1968]], device='cuda:0')
>>>

安装完毕

三、安装cocoapi

教程地址：https://github.com/philferriere/cocoapi
点击网址，打开readme.md，根据链接找到Visual C++ 2015

点击默认安装即可
在线安装后就能安装pycocotool了

pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

如果pip安装可能会出现问题，这里提供另外一种方法，直接去https://github.com/philferriere/cocoapi下载项目，然后在anaconda的激活环境中，我是根据这个方法进行安装的，进到cocoapi\PythonAPI目录：

(maskrcnn_pytorch) C:\WINDOWS\system32>d:

(maskrcnn_pytorch) D:\>cd D:\ChromeCoreDownloads\cocoapi-master\cocoapi-master\PythonAPI

(maskrcnn_pytorch) D:\ChromeCoreDownloads\cocoapi-master\cocoapi-master\PythonAPI>

执行下列命令：

# install pycocotools locally
python setup.py build_ext --inplace

# install pycocotools to the Python site-packages
python setup.py build_ext install

注意，＃为注释，可不用写在命令行中，执行完毕后完成安装
在D:\ChromeCoreDownloads\cocoapi-master\cocoapi-master\PythonAPI\pycocotools中，可以看见增加文件：

安装完成之后python命令进入python环境，能运行from pycocotools.coco import COCO和import pycocotools._mask as _mask即大功告成。

(maskrcnn_pytorch) C:\WINDOWS\system32>python
Python 3.7.11 (default, Jul 27 2021, 09:42:29) [MSC v.1916 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> from pycocotools.coco import COCO
>>> import pycocotools._mask as _mask
>>>

PS：如果有github项目中有pycocotools文件夹则import pycocotools的时候会优先import当前目录下的文件，如果在命令行中import没有问题在jupyter notebook上发现“No module named ‘pycocotools._mask”等问题，则只需把上文生成的pycocotools文件夹替换掉当前github项目目录下的pycocotools文件夹即可

四、安装深度学习加速库apex

克隆项目：https://github.com/NVIDIA/apex
cd到该项目文件夹，执行指令：

(maskrcnn_pytorch) C:\WINDOWS\system32>d:

(maskrcnn_pytorch) D:\>cd D:\ChromeCoreDownloads\apex-master\apex-master

(maskrcnn_pytorch) D:\ChromeCoreDownloads\apex-master\apex-master>

输入指令

pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

此时会有很多的信息跳出，这里不再记录，安装完毕后，进入python，输入

from apex import amp

具体的步骤可以参考这篇博客https://blog.csdn.net/qq_34914551/article/details/103203862
最后，再安装apex的相关库，再当前文件夹下运行

python setup.py install

出现

Installed d:\anaconda\program\envs\maskrcnn_pytorch\lib\site-packages\apex-0.1-py3.7.egg
Processing dependencies for apex==0.1
Finished processing dependencies for apex==0.1

完成安装

五、安装maskrcnn benchmark

cd 到下载好的maskrcnn-benchmark 目录下

python setup.py build develop

完成安装

六、配置pycharm

打开pycharm，打开程序

配置解释器，点击File->setting->project，可以看到当前的环境：

点击右上角的齿轮，选择Add…，出现如下界面：

选择Existing Enviroment,找到本地虚拟环境的安装地址
我的是D:\Anaconda\program\envs\maskrcnn_pytorch\python.exe
点击OK，发现已经设置成功。

至此pycharm环境配置完毕。

七、配置jupyter notebook

anaconda会自带jupyter notebook，但是notebook默认运行的是base环境，因此当我们创建其他虚拟环境时启动notebook需要另行设置。
要想在jupyter中使用刚刚创建的虚拟环境，需要进入刚才的虚拟环境中安装ipykernel：

pip install ipykernel
conda install jupyter
conda install nb_conda

接着修改配置文件（最重要的一步），进入Anaconda安装目录：我的目录为D:\Anaconda\program\share\jupyter\kernels，在里面新建一个文件夹，文件夹命名为虚拟环境的名字maskrcnn_pytorch，找到虚拟环境的对应位置，我的是D:\Anaconda\program\envs\maskrcnn_pytorch\share\jupyter\kernels\python3，将python3文件夹里的三个文件全部复制到新文件夹去

打开kernel.json并对"display_name"进行修改：

{
 "argv": [
  "D:/Anaconda/program/envs/maskrcnn_pytorch\\python.exe",#这是虚拟环境的目录
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "display_name": "maskrcnn_pytorch",# 这是在jupyter里看见的虚拟环境的名字，可以和创建的虚拟环境名称保持一致
 "language": "python"
}

在左下角Terminal中输入：

Jupyter-notebook

同时，浏览器中也会打开Jupyter-notebook

打开demo文件夹中的Mask_R-CNN_demo.ipynb

选择建立的虚拟环境即可。
其余的方法也可以直接通过

conda install nb_conda

详细请看：https://www.jianshu.com/p/afea092dda1d

总结

Mask R-CNN的环境配置比较复杂，建议从头建立一个虚拟环境，方便以后运用在自己的数据集上。
相关参考：https://blog.csdn.net/qq_38732601/article/details/106450549?ops_request_misc=&request_id=&biz_id=102&utm_term=maskrcnn%20%E7%8E%AF%E5%A2%83%20torch&utm_medium=distribute.pc_search_result.none-task-blog-2_allsobaiduweb~default-1-.first_rank_v2_pc_rank_v29&spm=1018.2226.3001.4187