在windows系统上安装caffe

下载编译

0.确认电脑上有VS2013

0.确认显卡GPU Compute Capability>=3.0

1.安装CUDA7.5

2.下载cuDNN v4，添加到CUDA7.5

3.根据https://github.com/Microsoft/caffe进行编译（64位Release模式）

4.需要下载的附加包已传到百度云NugetPackages与caffe文件夹并列存放

获取和生成caffe使用的Mnist数据集

由于自带的脚本是针对Linux系统的，需要修改

get_mnist.sh1.bat

echo "Downloading..."

set wget="../../../3rdparty/tools/wget.exe"

for %%i in (train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte) do %wget% --no-check-certificate http://yann.lecun.com/exdb/mnist/%%i.gz 

echo "done"

get_mnist.sh2.bat

echo "Renaming..."

set do_7za="../../../3rdparty/tools/7za.exe"

for %%i in (train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte) do %do_7za% x %%i.gz

rename train-images.idx3-ubyte train-images-idx3-ubyte
rename train-labels.idx1-ubyte train-labels-idx1-ubyte
rename t10k-images.idx3-ubyte t10k-images-idx3-ubyte
rename t10k-labels.idx1-ubyte t10k-labels-idx1-ubyte

echo "done"

create_mnist-lmdb.sh.bat

set DATA=../../data/mnist
set EXAMPLE=../../examples/mnist
set TOOLS=../../Build/x64/Release

set BACKEND=lmdb
REM set BACKEND=leveldb

echo "Creating %BACKEND%..."

rd /s /q "mnist_train_%BACKEND%"
rd /s /q "mnist_test_%BACKEND%"

"%TOOLS%/convert_mnist_data.exe" %DATA%/train-images-idx3-ubyte %DATA%/train-labels-idx1-ubyte mnist_train_%BACKEND% --backend=%BACKEND%
"%TOOLS%/convert_mnist_data.exe" %DATA%/t10k-images-idx3-ubyte %DATA%/t10k-labels-idx1-ubyte mnist_test_%BACKEND% --backend=%BACKEND%

echo "Done."

pause

train_lenet.sh.bat

cd ../../
"Build/x64/Release/caffe.exe" train --solver=examples/mnist/lenet_solver.prototxt
pause

测试结果

python支持

1.安装anaconda

2.cmd运行pip install protobuf

3.修改CommonSettings.props然后生成pycaffe项目

4.添加环境变量，“PythonPath” 指向相应路径Buildx64Releasepycaffe

5.import caffe无报错即通过

matlab支持

1.安装matlab

2.修改CommonSettings.props然后生成matcaffe项目

3.将相应路径Buildx64Release添加到path环境变量

4.把相应路径Buildx64Releasematcaffe添加到matlab的search path中

5.运行classification_demo.m

>> classification_demo
using caffe/examples/images/cat.jpg as input image
Elapsed time is 0.078070 seconds.
Elapsed time is 0.381840 seconds.
Cleared 0 solvers and 1 stand-alone nets