WINDOW10 System Caffe installation and configuration Matlab interface (video card is 1080)

Caffe is generally installed under the Linux system, online about the installation of Windows Tutorial tutorial, and each tutorial is not very full, I am here to summarize the process and all the solutions to the bug.

I am the Win10+gtx1080+vs2013,matlab interface I am matlab2016a.

1. Install Visual Studio 2013 first. It's not much of a difficulty, just download and install it online.

2. Download Caffe.https://github.com/microsoft/caffe, click Clone or download-> download ZI

Unzip, open to the Caffe-master/caffe-master/windows folder.

Copy the CommonSettings.props.example file and rename the copied file to Commonsettings.props.

Open the Caffe.sln file with VS2013 and click on the Commonsettings.props on the right to see the version of your CuDNN support

3. Installing CUDA and CuDNN
Open CUDA and pull below to click DOWNLOAD

Select Windows-x86_64-10-exe, if your operating system is a different version, make a selection in version

Click Run after download, install path Please select default path

Open CuDNN, click Download, before downloading need to register

After registration is complete, download CuDNN v4 or CuDNN v5 (depending on the version supported in the Commonsettings.props file, it is recommended to select CuDNN v4 Library for Windows, although for CUDA7.0)

4. Prepare before compiling

Open to Caffe-master/caffe-master/windows folder, run Caffe.sln file with VS2013

The right side will show Libcaffe (failed to load), the steps to resolve:

1) Open the Msbuildextensions folder in the CUDA installation path, if you selected the default path at the time of installation, then it should be in the C drive/program File/nvidia GPU Cpmputingtoolkit/cuda/8.0/ex Tras/visual_studio_integration.

2) Copy all the files in the Msbuildextensions folder to the C Drive/program file (x86)/msbuild/microsoft.cpp/v4.0 (depending on the version you installed)/v120/ Buildcustomizations folder, replace the files in the target

3) Click on the Commonsettings.props file on the right of VS2013.

Line Seventh Cpuonlybuild If True is the CPU version, False is the GPU version, the GPU version below the eighth line to use CUDNN to True,cuda version to our version 8.0.

The following pythonsupport and Matlabsupport are the options for which interfaces are supported, both in Python and Matlab, and two to true.

Line 24th to the path of the Cuda folder in your CuDNN decompression file

The 53rd line to your MATLAB installation folder location, 55 lines to increase the path shown in the figure,

Otherwise, there's a problem with the GPU on the back of the compilation.

On the keyboard control + s save, close VS2013 and reopen the Caffe.sln file, then Libcaffe should be loaded successfully

5. Compiling Caffe

Open Caffe.sln file, select Debug-Caffe Property

Click Configuration Properties, above configuration selected as Active (Release), platform selected as active (x64)

Then click Local debugging to start compiling, which will automatically download the NuGet package.

Of course, but download the problem, you can directly open options, Package Manager, General---Browse,

Manually copy the NuGet package in, the password Qfui

Error LNK1104: Could not open file "Libcaffe.lib"

Error C2220: Warning is treated as error-no "object" file generated

The solution is to regenerate the Libcaffe separately and make the following modifications to the Libcaffe property

If there is a problem with OpenCV during compilation, double-click the Error statement to comment out the part about OpenCV, and regenerate Libcaffe should be possible.

This should compile successfully, and finally click Build-Build solution should be able to build successfully.

When this file appears, it shows success.

6. Test whether MATLAB and Python are available caffe

Add Matcaffe from the Caffe file directory to the path of MATLAB

Caffe-master\caffe-master\build\x64\release\matcaffe, click the Set path in the menu, you can call the Caffe library in MATLAB every time.

and add the path of the release to the system environment variable in the path, restart the computer.

Open matlab and test it.

Indicates that the MATLAB interface can be used.

Refer to these two articles specifically

http://forestli.com/archives/93/

Http://www.cnblogs.com/yixuan-xu/p/5858595.html

WINDOW10 System Caffe installation and configuration Matlab interface (video card is 1080)