Windows environment Caffe Installation configuration steps (no GPU) and mnist training

In the second year of Master's degree, he threw himself into the wave of deep learning. From the previous inertial navigation to this direction, everything starts from the beginning, here, only in this article to record their own way of playing strange.

The initial idea is to get familiar with Caffe, considering the difficulty of getting started with the Ubuntu , so start with the basics in Windows. There is an episode in which the desktop can only be maintained on 32-bit systems for some reason, the system is not compatible when compiling caffe.cpp, and then it is switched to 64-bit notebooks for operation.

Pre-Preparation: 1. VS 2013 2. Windows version of Caffe (https://github.com/BVLC/caffe/tree/windows), directly download.

Caffe Installation:

Step1

Unzip the Caffe from GitHub, copy the CommonSettings.props.example file from the./windows into this folder, and then delete example get the Commonsettings.props file,

If you do not do this, you will not open the CommonSettings.props.example file in vs.

Step2

Modify the Commonsettings.props file. Modify the status according to your own needs, such as my own notebook without GPU, in the corresponding place to set the bit false, corresponding, I can only use the CPU, then changed to True;matlab and Python interface is enabled (enabled) is also set here.

Problems that may occur: Libcaffe and Test_all failed to load; Workaround: Unzip the package from the beginning of the first step.

Step3

Build the solution and download the third-party library.

Open the./windows in Caffe.sln, right-click solution "Caffe", tap Build Solution

At this time, a box will pop up, indicating that the library is under the third party, including OpenCV and other libraries, can be found after downloading

The problem arises: the build solution failed, prompting that error such as Libcaffe.lib could not be opened. The workaround is to regenerate the Libcaffe directly, or generate it directly once

Then there is the problem: the object file is not generated; The workaround: Double-click the error, save the pop-up box, and then Libcaffe "Regenerate once"

Finally, the solution build is successful.

Step 4

Setting properties

STEP5

Run Caffe.cpp, press CTRL+F5 to compile

This situation occurs:

The solution is to restart the computer.

The following window appears to illustrate the success of the compilation

Mnist Test

can refer to http://blog.csdn.net/qq_14845119/article/details/52415090

It is important to note that:

1. Download them all and set up two folders in the./data/mnist to install each one

2. scripting language is to check whether the path is correct, to use their own path, can not copy the author.

3. If CPU only, the network training time is about one hour (may be my notebook configuration low)

In this special thanks to the "wind-wing Ice Boat" http://blog.csdn.net/zb1165048017/article/details/51355143, important ideas from this article

First write the text, mistakes inevitably, please correct me! If infringement of your intellectual property rights, please contact me, e-mail: [Email protected]

Windows environment Caffe Installation configuration steps (no GPU) and mnist training