Generate Caffe under windows+vs2013 and perform CIFAR10 classification test

http://blog.csdn.net/naaaa/article/details/52118437Tags: windowsvs2013caffecifar102016-08-04 15:33 1316 People read Comments (1) favorite reports Classification:Caffe

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

1. Download vs2013, install

Http://download.microsoft.com/download/0/7/5/0755898A-ED1B-4E11-BC04-6B9B7D82B1E4/VS2013_RTM_ULT_CHS.iso

2. Download Caffe source code, unzip

Https://github.com/Microsoft/caffe

Remove the. Example behind the Caffe-master/windows CommonSettings.props.example.

3.GPU Configuration

Configuration with GPU:

Download Cuda:

Https://developer.nvidia.com/cuda-downloads

Select the downloaded version according to your video card

Double-click EXE to install

Download CUDNN:

Https://developer.nvidia.com/cudnn

Download V4 or V5 version

After decompression, respectively, the LIB, include, bin folder files are copied to the Cuda installation directory under the Lib, include, bin folder

(Default: C:\Program files\nvidia GPU Computing toolkit\cuda\vx.x)

GPU-Free configuration:

Notepad to open caffe-master/windows under Commonsettings.prop Modify:

<CpuOnlyBuild>false</CpuOnlyBuild> for <CpuOnlyBuild>true</CpuOnlyBuild>

<UseCuDNN>true</UseCuDNN> for <UseCuDNN>false</UseCuDNN>

To compile each project before setting:

Properties---Property---Configuration Properties->c/c++-> preprocessor---preprocessor definitions, adding an item cpu_only

4. Generate Libcaffe

Open the Caffe.sln under Caffe-mastetr/windows

All projects need to be libcaffe.lib, so the first one to compile this project.

Libcaffe has a lot of relevant libraries, and the project has been configured to download with NuGet.

We need to install the NuGet Package Manager and start it.

Download NuGet:

Tools, extensions and updates, online, search nuget, download NuGet package management

(You can also download Http://docs.nuget.org/consume/installing-nuget website directly, double-click Install)

Start NuGet:

Right-click Project, enable NuGet package restore.

The resulting Libcaffe will be generated under Caffe-master\build\x64\debug libcaffe.lib

(all subsequent EXE files are also generated in this directory)

5.CIFAR10 Training Data

Http://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz

Download data to Caffe-master\data\cifar10, unzip.

The downloaded data is in binary format and needs to be converted to LEVELDB.

Compile the Convert_cifar_data project and generate the Convert_cifar_data.exe.

Open CMD,CD to Caffe-master\build\x64\debug and enter the command (all subsequent commands are under this path):

Convert_cifar_data.exe. /.. /.. /data/cifar10/cifar-10-batches-bin. /.. /.. /data/cifar10 Leveldb

Generate Cifar10_test_leveldb and Cifar10_train_leveldb folders under the Caffe-master\data\cifar10 folder

6. Calculate the data image mean value

Compile Compute_image_mean project, generate Compute_image_mean.exe

CMD input command:

Compute_image_mean.exe. /.. /.. /data/cifar10/cifar10_train_leveldb. /.. /.. /data/cifar10/mean.binaryproto--backend=leveldb

7. Training CIFAR10 Model

Open Caffe-master\examples\cifar10\cifar10_quick_solver.prototxt File:

Modify path: Examples/cifar10 for. /.. /.. /examples/cifar10 (two places)

Modify the last behavior: solver_mode:cpu (do not change this one with the GPU)

Open caffe-master\examples\cifar10\ Cifar10_quick_train_test.prototxt file:

Modified Backend:lmdb to Backend:leveldb (two places)

Modify Mean_file: "Examples/cifar10/mean.binaryproto" for Mean_file: ". /.. /.. /data/cifar10/mean.binaryproto "(Two places)

Modify Source: "Examples/cifar10/cifar10_train_lmdb" as Source: ". /.. /.. /data/cifar10/cifar10_train_leveldb "

Modify Source: "Examples/cifar10/cifar10_test_lmdb" as Source: ". /.. /.. /data/cifar10/cifar10_test_leveldb "

Compile Caffe project, generate Caffe.exe

CMD input command: Caffe.exe train--solver=. /.. /.. /examples/cifar10/cifar10_quick_solver.prototxt Training Network

CPU training will be trained for a long time, after the completion of training Caffe-master/examples/cifar10 folder generated

Cifar10_quick_iter_4000.caffemodel.h5

Cifar10_quick_iter_4000.solverstate.h5

8. Classification of images

Create a new text file Synset_words.txt under Caffe-master\data\cifar10.

The content of the file is the category of classification contained in Cifar10, as follows:

[Plain]View PlainCopyprint?

1. Airplane
2. Automobile
3. Bird
4. Cat
5. Deer
6. Dog
7. Frog
8. Horse
9. Ship
10. Truck

Compile classification project, generate Classification.exe

Command Line Input:

Classification.exe. /.. /.. /examples/cifar10/cifar10_quick.prototxt. /.. /.. /examples/cifar10/cifar10_quick_iter_4000.caffemodel.h5. /.. /.. /data/cifar10/mean.binaryproto. /.. /.. /data/cifar10/synset_words.txt. /.. /.. /examples/images/cat.jpg

Will come out and classify the results, my results:

[Plain]View PlainCopy print?

1. ----------prediction for. /.. /.. /examples/images/cat.jpg----------
2. 0.9784-"Deer"
3. 0.0100-"Cat"
4. 0.0094-"Bird"
5. 0.0017-"Frog"
6. 0.0004-"Dog"

Reference post: http://blog.csdn.net/zb1165048017/article/details/51476516

Generate Caffe under windows+vs2013 and perform CIFAR10 classification test