转载来自://bbsmax.ikafan.com/static/L3Byb3h5L2h0dHAvYmxvZy5jc2RuLm5ldC9ndW95azE5OTAvYXJ0aWNsZS9kZXRhaWxzLzUyOTA5ODY0.jpg
本文主要介绍官方给出的caffe-windows的配置及如何训练mnist数据集,介绍的比较基础,大神请绕道
1、环境:windows 10\CUDA7.5\cuDNN\VS2013
2、GPU计算环境准备(没有GPU的同学可以跳过此步)
(1)首先下载并安装CUDA7.5,下载界面如图1:
图 1:CUDA7.5的下载界面
(2)下载cuDNN,注意要下载cuDNN v4版本,下载界面如图2:
图 2:cuDNN的下载界面
官网下载cuDNN需要先注册,而且要填一些调查表,也可以在我上传的资源中下载。下载后解压会有三个文件夹bin、include、lib。将这三个文件夹复制到cuda的安装目录中:\NVIDIA GPU ComputingToolkit\CUDA\v7.5。(cuda的安装目录中也有这三个文件夹,将这三个文件夹分别与原来存在的文件夹合并,如3图所示)。
图 3:CUDA 7.5 安装的根目录
3、caffe-windows准备
(1)下载官方caffe-windows并解压,将 .\windows\CommonSettings.props.example备份,并改名为CommonSettings.props。如图4所示:
图 4:修改后的CommonSettings.props文件
(2)关于CommonSettings.props文件的一点说明。
- </pre><pre name="code" class="html"><?xml version="1.0" encoding="utf-8"?>
- <Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
- <ImportGroup Label="PropertySheets" />
- <PropertyGroup Label="UserMacros">
- <BuildDir>$(SolutionDir)..\Build</BuildDir>
- <!--NOTE: CpuOnlyBuild and UseCuDNN flags can't be set at the same time.-->
- <CpuOnlyBuild>false</CpuOnlyBuild><!--注释里说的很清楚,这两个值不能同时设为true。若没有GPU就把CpuOnlyBuild设为true-->
- <UseCuDNN>true</UseCuDNN>
- <CudaVersion>7.5</CudaVersion>
- <!-- NOTE: If Python support is enabled, PythonDir (below) needs to be
- set to the root of your Python installation. If your Python installation
- does not contain debug libraries, debug build will not work. -->
- <PythonSupport>false</PythonSupport><!--设置是否支持python接口,若想支持,需要改后面的PythonDir的值-->
- <!-- NOTE: If Matlab support is enabled, MatlabDir (below) needs to be
- set to the root of your Matlab installation. -->
- <MatlabSupport>false</MatlabSupport><!--设置是否支持matlab接口,若想支持,需要改后面的MatlabDir的值-->
- <CudaDependencies></CudaDependencies>
- <!-- Set CUDA architecture suitable for your GPU.
- Setting proper architecture is important to mimize your run and compile time. -->
- <CudaArchitecture>compute_35,sm_35;compute_52,sm_52</CudaArchitecture>
- <!-- CuDNN 3 and 4 are supported -->
- <CuDnnPath></CuDnnPath>
- <ScriptsDir>$(SolutionDir)\scripts</ScriptsDir>
- </PropertyGroup>
- <PropertyGroup Condition="'$(CpuOnlyBuild)'=='false'">
- <CudaDependencies>cublas.lib;cuda.lib;curand.lib;cudart.lib</CudaDependencies>
- </PropertyGroup>
- <PropertyGroup Condition="'$(UseCuDNN)'=='true'">
- <CudaDependencies>cudnn.lib;$(CudaDependencies)</CudaDependencies>
- </PropertyGroup>
- <PropertyGroup Condition="'$(UseCuDNN)'=='true' And $(CuDnnPath)!=''">
- <LibraryPath>$(CuDnnPath)\cuda\lib\x64;$(LibraryPath)</LibraryPath>
- <IncludePath>$(CuDnnPath)\cuda\include;$(IncludePath)</IncludePath>
- </PropertyGroup>
- <PropertyGroup>
- <OutDir>$(BuildDir)\$(Platform)\$(Configuration)\</OutDir>
- <IntDir>$(BuildDir)\Int\$(ProjectName)\$(Platform)\$(Configuration)\</IntDir>
- </PropertyGroup>
- <PropertyGroup>
- <LibraryPath>$(OutDir);$(CUDA_PATH)\lib\$(Platform);$(LibraryPath)</LibraryPath>
- <IncludePath>$(SolutionDir)..\include;$(SolutionDir)..\include\caffe\proto;$(CUDA_PATH)\include;$(IncludePath)</IncludePath>
- </PropertyGroup>
- <PropertyGroup Condition="'$(PythonSupport)'=='true'"><!--与前面python接口设置对应-->
- <PythonDir>C:\Miniconda2\</PythonDir>
- <LibraryPath>$(PythonDir)\libs;$(LibraryPath)</LibraryPath>
- <IncludePath>$(PythonDir)\include;$(IncludePath)</IncludePath>
- </PropertyGroup>
- <PropertyGroup Condition="'$(MatlabSupport)'=='true'"><!--与前面的matlab接口设置对应-->
- <MatlabDir>C:\Program Files\MATLAB\R2014b</MatlabDir>
- <LibraryPath>$(MatlabDir)\extern\lib\win64\microsoft;$(LibraryPath)</LibraryPath>
- <IncludePath>$(MatlabDir)\extern\include;$(IncludePath)</IncludePath>
- </PropertyGroup>
- <ItemDefinitionGroup Condition="'$(CpuOnlyBuild)'=='true'">
- <ClCompile>
- <PreprocessorDefinitions>CPU_ONLY;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- </ClCompile>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup Condition="'$(UseCuDNN)'=='true'">
- <ClCompile>
- <PreprocessorDefinitions>USE_CUDNN;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- </ClCompile>
- <CudaCompile>
- <Defines>USE_CUDNN</Defines>
- </CudaCompile>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup Condition="'$(PythonSupport)'=='true'">
- <ClCompile>
- <PreprocessorDefinitions>WITH_PYTHON_LAYER;BOOST_PYTHON_STATIC_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- </ClCompile>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup Condition="'$(MatlabSupport)'=='true'">
- <ClCompile>
- <PreprocessorDefinitions>MATLAB_MEX_FILE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- </ClCompile>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup>
- <ClCompile>
- <MinimalRebuild>false</MinimalRebuild>
- <MultiProcessorCompilation>true</MultiProcessorCompilation>
- <PreprocessorDefinitions>_SCL_SECURE_NO_WARNINGS;USE_OPENCV;USE_LEVELDB;USE_LMDB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- <TreatWarningAsError>true</TreatWarningAsError>
- </ClCompile>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
- <ClCompile>
- <Optimization>Full</Optimization>
- <PreprocessorDefinitions>NDEBUG;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
- <FunctionLevelLinking>true</FunctionLevelLinking>
- </ClCompile>
- <Link>
- <EnableCOMDATFolding>true</EnableCOMDATFolding>
- <GenerateDebugInformation>true</GenerateDebugInformation>
- <LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
- <OptimizeReferences>true</OptimizeReferences>
- </Link>
- </ItemDefinitionGroup>
- <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
- <ClCompile>
- <Optimization>Disabled</Optimization>
- <PreprocessorDefinitions>_DEBUG;%(PreprocessorDefinitions)</PreprocessorDefinitions>
- <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
- </ClCompile>
- <Link>
- <GenerateDebugInformation>true</GenerateDebugInformation>
- </Link>
- </ItemDefinitionGroup>
- </Project>
4、编译caffe-windows
编译用vs2013打开.\windows\Caffe.sln 并将解决方案的配置改为release,点菜单栏上的“生成->生成解决方案”,会将整个项目全部生成,这个时间会比较长(由于官方caffe-windows 的版本使用了NuGet管理第三方开发包,所以需要在vs2013上安装NuGet,官方网站下载速度比较慢,可以在我的资源里下载)。生成成功之后的文件都在.\Build\x64\Release中。
PS:生成时可能遇到的错误:errorC2220: 警告被视为错误 - 没有生成“object”文件 (..\..\src\caffe\util\math_functions.cpp)。这个错误可参考Sunshine_in_Moon 的解决方案。
5、测试
1)下载MNIST数据集,MNIST数据集包含四个文件,如表1所示:
表1:MNIST数据集及其文件解释
文件 | 内容 |
训练集图片 - 55000 张 训练图片, 5000 张 验证图片 | |
训练集图片对应的数字标签 | |
测试集图片 - 10000 张 图片 | |
测试集图片对应的数字标签 |
2)转换 训练\测试数据
a) 中的四个文件放到 . \examples\mnist\mnist_data文件夹下。
b) 在caffe-windows安装的根目录下,新建一个convert-mnist-data-train.bat文件转换为训练数据,并在文件中添加代码:
- Build\x64\Release\convert_mnist_data.exe --backend=lmdbexamples\mnist\mnist_data\train-images.idx3-ubyteexamples\mnist\mnist_data\train-labels.idx1-ubyte examples\mnist\mnist_data\mnist_train_lmdb
- pause
其中--backend=lmdb 表示转换为lmdb格式,若要转换为leveldb将其改写为--backend=leveldb 即可。
再新建一个convert-mnist-data-test.bat转换测试数据,代码为:
- Build\x64\Release\convert_mnist_data.exe --backend=lmdb examples\mnist\mnist_data\t10k-images.idx3-ubyte examples\mnist\mnist_data\t10k-labels.idx1-ubyte examples\mnist\mnist_data\mnist_test_lmdb
- Pause
Ps:(1)convert_mnist_data.exe的命令格式为:
convert_mnist_data [FLAGS] input_image_file input_label_file output_db_file
[FLAGS]:转换的文件格式可取leveldb或lmdb,示例:--backend=leveldb
Input_image_file:输入的图片文件,示例:train-images.idx3-ubyte
input_label_file:输入的图片标签文件,示例:train-labels.idx1-ubyte
output:保存输出文件的文件夹,示例:mnist_train_lmdb
(2)如果感觉很麻烦,也可以直接下载我转换好的MNIST文件(leveldb和lmdb)。
3)运行测试
(1)将第2)步中转换好的训练\测试数据集(mnist_train_lmdb\ mnist_train_lmdb或mnist_train_leveldb\mnist_train_leveldb)文件夹放在.\examples\mnist中。
(2)在caffe-windows根目录下新建一个run.bat,文件中代码:
- Build\x64\Release\caffe.exe train --solver=examples/mnist/lenet_solver.prototxt
- pause
保存并双击运行,如果运行成功,说明caffe配置成功了。
注意:使用leveldb或lmdb格式的数据时,需要将lenet_train_test.prototxt 文件里面的data_param-> source和data_param-> backend相对应,如图5红框所标注处。
图 5:lenet_train_test.prototxt文件中需要注意与训练\测试数据对应的部分
4)训练自己的数据
这部分可以参考下面的几个博客: