caffe-Cuda7.5-cudnnv4-GPU-NugetPackages-Tesla k40-VS2013-Anaconda2-pycharm2016.2 win10

1.caffe  Cuda7.5  cudnnv4  GPU  NugetPackages Tesla k40  VS2013  Anaconda2  pycharm2016.2    win10 

2.各个文件夹:

下载链接:链接:http://pan.baidu.com/s/1jIwnA1w 密码:g6oy

  caffe: F:\caffe\caffe-master
  cudnn: F:\caffe\cuda
  Cuda7.5: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5
  Anaconda: D:\Anaconda2
  NugetPackages: F:\caffe\NugetPackages

3. cudnn文件下cuda\bin下cudnn64_4.dll拷贝到C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin文件夹下。同理cuda\include和cuda\lib\x64文件夹下文件拷贝到CUDA\v7.5对应的include和lib\x64文件夹下。
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\extras\visual_studio_integration\MSBuildExtensions文件夹下文件拷贝到C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V120\BuildCustomizations文件夹下

4.环境变量:
  添加新的系统环境变量:
  变量名:CuDnnPath
  值:F:\caffe
  
  系统变量Path中编辑添加:
  F:\caffe\cuda\bin
  C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin
  F:\caffe\caffe-master\Build\x64\Release

5.复制F:\caffe\caffe-master\windows\CommonSettings.props.example粘贴到此文件夹下之后重命名为CommonSettings.props。用vs2013打开Caffe.sh。

6.修改之后的内容如下:
.........
 <CpuOnlyBuild>false</CpuOnlyBuild>
 <UseCuDNN>true</UseCuDNN>
 <CudaVersion>7.5</CudaVersion>
               ..........
 <PythonSupport>true</PythonSupport>
..........
 <CudaArchitecture>compute_35,sm_35</CudaArchitecture>
........
 <!-- CuDNN 4 and 5 are supported -->
        <CuDnnPath>F:\caffe</CuDnnPath>
.........
   <PythonDir>D:\Anaconda2\</PythonDir>

## 这里Tesla k40计算能力3.5所以compute_35,sm_35(查询GPU计算能力wiki链接:https://en.wikipedia.org/wiki/CUDA)

7.右击libcaffe-Set as StartUp Project
  工具栏-DEBUG-libcaffe Properties-选编译Release
  右击libcaffe-Build
......
  1 succeeded ,0 failed

8.pycaffe-Set as StartUp Project  Release

9.  VC++:
D:\Anaconda2\include
D:\Anaconda2\Lib\site-packages\numpy\core\include
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\include
D:\Anaconda2\Lib

D:\Anaconda2\libs
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\lib\x64
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\lib

C/C++:
include:
F:\caffe\cuda\include
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\include
D:\Anaconda2\include
D:\Anaconda2
D:\Anaconda2\Lib
D:\Anaconda2\Lib\site-packages\numpy\core\include

Linker-Input-Additional Dependenceies:
F:\caffe\cuda\lib\x64
F:\caffe\cuda\lib
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\lib\x64
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\include
D:\Anaconda2\libs
D:\Anaconda2\Lib\site-packages\numpy\core\lib

10.pycaffe-build
   succeeded

12.Path:F:\caffe\caffe-windows\python
   将F:\caffe\caffe-windows\Build\x64\Release\pycaffe下caffe文件夹复制到D:\Anaconda2\Lib\site-packages下即可

13.测试Demo:caffe-test.py
   注意网盘文件Demo下bvlc_reference_caffenet.caffemodel要拷贝到F:\caffe\caffe-master\models\bvlc_reference_caffenet文件夹下
  synset_words.txt要拷贝到F:\caffe\caffe-master\data\ilsvrc12文件夹下。

caffe.set_device(0)     #模式切换
caffe.set_mode_gpu()
#caffe.set_mode_cpu()

部分错误解决办法:
错误:error C1083: Cannot open source file: ‘..\..\src\caffe\data_reader.cpp‘
项目引用了已经不存在的源文件,只要在项目include和src中删除data_reader的索引即可。

如果需要安装protobuf:
 添加环境变量  ...\protobuf\python(重启)
1.下载protobuf(地址:https://github.com/google/protobuf/releases/tag/v3.0.0),下载两个版本,一个protoc-3.0.0-win32.zip,一个源码。
2.将protoc-3.0.0-win32\bin\protoc.exe 拷贝进入源代码文件夹下 src中
3、进入源代码文件夹下Python文件夹,cmd执行 python setup.py build,这里我出现了ImportError: No module named setuptools,解决方案(http://blog.sina.com.cn/s/blog_3fe961ae0100zgav.html)
4.重新进入源代码文件夹下python文件夹,cmd执行 python setup.py build、执行 python setup.py install

如果需要安装opencv:
import cv2时会出现这个问题
解决方法:将OpenCV安装目录里的Python文件夹内的cv2.pyd复制到Python安装目录里Lib中site-packages内即可解决

cmd下nvidia-smi查看GPU使用设备情况

CPU配置去掉和CUDA、cudnn所有相关的即可


猜你喜欢

转载自blog.csdn.net/md2017/article/details/60323346
今日推荐