[Win10] mmcv 和 mmdetection 安装过程图文详解

环境记录

• 笔记本: MSI GE63 Raider RGB 8SG-040CN
• CPU: i7-8750H
• GPU: RTX 2080
• 系统: Win10 专业版
• 显卡驱动: 496.76
• Python: 3.7.13
• Pytorch: 1.10.1
• CUDA: 10.2
• cuDNN: 8.3.1
• MMCV: 1.5.1
• mmdetection: 2.24.1

Step1: 环境配置

安装Anaconda, CUDA, cuDNN

教程很多，不细说了，假设已经能跑成功PyTorch代码了。

创建Conda环境

conda create --name mmdet python=3.7
conda activate mmdet 

安装PyTorch

conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=10.2 -c pytorch

查看CUDA版本：

nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Wed_Oct_23_19:32:27_Pacific_Daylight_Time_2019
Cuda compilation tools, release 10.2, V10.2.89

查看cuDNN版本：
进入C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\include
找到cudnn_version.h打开就可以看到

#ifndef CUDNN_VERSION_H_
#define CUDNN_VERSION_H_

#define CUDNN_MAJOR 8
#define CUDNN_MINOR 3
#define CUDNN_PATCHLEVEL 1

cuDNN版本即为8.3.1

安装MSVC编译器

如果是CUDA10.2的话，这里只能安装Visual Studio 2015到Visual Studio 2019之间的版本，我这里是安装的Visual Studio 2019 下载链接 安装Community版就行。
如果是CUDA10.2的话
Visual Studio 2022 亲测不行，后面编译会报错！！！
Visual Studio 2022 亲测不行，后面编译会报错！！！
Visual Studio 2022 亲测不行，后面编译会报错！！！

安装好后配置环境变量：

找到Visual Studio安装文件夹，我这里是在E盘安装的：

找到并进入VC文件夹

找到并进入Tools文件夹：

进入MSVC文件夹

进入14.29.30133文件夹，这里因版本而异，数字可能不一样，点进去就行

进入bin文件夹

进入Hostx86文件夹

进入x64文件夹

找到cl.exe

把之前的文件夹路径加入到环境变量里：
例如我的路径是

E:\VisualStudio19\VC\Tools\MSVC\14.29.30133\bin\Hostx86\x64

加入系统Path中

注：默认安装Visual Studio的话路径是

C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.27.29110\bin\Hostx86\x64

测试

命令行中输入cl，输入如下即为配置成功。

C:\Users\Leo>cl
用于 x64 的 Microsoft (R) C/C++ 优化编译器 19.29.30145 版
版权所有(C) Microsoft Corporation。保留所有权利。

用法: cl [ 选项... ] 文件名... [ /link 链接选项... ]

安装mmcv-full

从Github上 clone 下来

conda activate mmdet
git clone https://github.com/open-mmlab/mmcv.git
# git checkout v1.2.0 # 如果你需要装特定版本
cd mmcv
pip3 install -r requirements.txt

查TORCH_CUDA_ARCH_LIST

命令行cd到C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\extras\demo_suite
运行deviceQuery.exe

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\extras\demo_suite>deviceQuery.exe
deviceQuery.exe Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "NVIDIA GeForce RTX 2080"
  CUDA Driver Version / Runtime Version          11.5 / 10.2
  CUDA Capability Major/Minor version number:    7.5
  Total amount of global memory:                 8192 MBytes (8589606912 bytes)
  (46) Multiprocessors, ( 64) CUDA Cores/MP:     2944 CUDA Cores
  GPU Max Clock rate:                            1590 MHz (1.59 GHz)
  Memory Clock rate:                             7001 Mhz
  Memory Bus Width:                              256-bit
  L2 Cache Size:                                 4194304 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               zu bytes
  Total amount of shared memory per block:       zu bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  1024
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          zu bytes
  Texture alignment:                             zu bytes
  Concurrent copy and kernel execution:          Yes with 2 copy engine(s)
  Run time limit on kernels:                     Yes
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  CUDA Device Driver Mode (TCC or WDDM):         WDDM (Windows Display Driver Model)
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      No
  Device PCI Domain ID / Bus ID / location ID:   0 / 1 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 11.5, CUDA Runtime Version = 10.2, NumDevs = 1, Device0 = NVIDIA GeForce RTX 2080
Result = PASS

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\extras\demo_suite>

信息第二行的这个数就是TORCH_CUDA_ARCH_LIST要填的数

CUDA Capability Major/Minor version number:    7.5

编译安装mmcv-full

打开powershell，输入以下内容

#             ---设置MMCV要用的一些环境变量---
$env:MMCV_WITH_OPS = 1
$env:MAX_JOBS = 8 # 根据你可用CPU以及内存量进行设置
$env:TORCH_CUDA_ARCH_LIST="7.5" #这个因显卡而异，下面我会说怎么查
#-----------------------------------------------------------------
#---设置CUDA_HOME或者CUDA_PATH，如果之前装CUDA的时候弄过就不用写了---
$env:CUDA_HOME = "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\"
#或者
$env:CUDA_HOME = $env:CUDA_PATH_V10_2 # 如果 CUDA_PATH_V10_2 在CUDA安装#过程中已经设置好
#-----------------------------------------------------------------
conda activate mmdet # 激活环境
cd mmcv # 改变路径
python setup.py build_ext # 如果成功, cl 将会自动弹出来编译 flow_warp
python setup.py develop # 安装
pip show mmcv-full # 检查是否安装成功
pip list # 检查是否安装成功

注意TORCH_CUDA_ARCH_LIST这一项因机器而异，按照下面的内容查完再填

我的cmd不能直接输入类似$env:MMCV_WITH_OPS = 1这种命令，会报文件名、目录名或卷标语法不正确。

python setup.py build_ext和python setup.py develop耗时较长，耐心等待即可

验证

最后pip show完之后应该是这样的 只有mmcv-full没有mmcv

(mmdet) E:\mmcv>pip show mmcv-full
Name: mmcv-full
Version: 1.5.1
Summary: OpenMMLab Computer Vision Foundation
Home-page: https://github.com/open-mmlab/mmcv
Author: MMCV Contributors
Author-email: [email protected]
License: UNKNOWN
Location: e:\mmcv
Requires: addict, numpy, packaging, Pillow, pyyaml, yapf, regex
Required-by:

(mmdet) E:\mmcv>pip show mmcv
WARNING: Package(s) not found: mmcv

Step2：安装MMdetection

命令行输入：

pip install mmdet

即可

Step3：验证

命令行输入：

conda activate mmdet
git clone https://github.com/open-mmlab/mmdetection.git
cd mmdetection
pip install openmim
mim download mmdet --config faster_rcnn_r50_fpn_1x_coco --dest . #下载模型

弄好之后会下载一个faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
之后创建一个文件夹checkpoints，把上面的模型放到./checkpoints下

最后运行如下代码进行验证

from mmdet.apis import init_detector, inference_detector
import mmcv
config_file = 'configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py'
# download the checkpoint from model zoo and put it in `checkpoints/`
# url: https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth
checkpoint_file = 'checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth'
device = 'cuda:0'
# init a detector
model = init_detector(config_file, checkpoint_file, device=device)
# inference the demo image
result = inference_detector(model, 'demo/demo.jpg')
outImg=model.show_result(
'demo/demo.jpg',
result)
mmcv.imshow(outImg,'a')
print(result)

如果出现如下结果则安装成功：

踩坑/报错分享区

1. 在CUDA 10.2版本下，安装Visual Studio 2022后，运行python setup.py build_ext会报错，显示xxx is in
2. 一开始参考的 https://zhuanlan.zhihu.com/p/308281195 安装，之后不知道是1的原因，因此在评论区找到了 这个链接：https://www.zywvvd.com/notes/environment/cuda/windows-mmcv-1-3-13-install/windows-mmcv-1-3-13-install/ 进行安装，但是最后发现这样弄只能安装上mmcv而非mmcv-full，最后验证的时候会报ModuleNotFoundError: No module named 'mmcv._flow_warp_ext'
3. 之后也尝试了https://mmdetection.readthedocs.io/en/v2.21.0/get_started.html里面的方法：

pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

运行python setup.py build_ext会报错，可能也是1的原因，这个我没去验证。

参考链接

• 环境配置参考 https://zhuanlan.zhihu.com/p/308281195 和 https://mmdetection.readthedocs.io/en/v2.21.0/get_started.html