First say nonsense: In the tutor's office 2 days + oneself at home Day finally completed Caffe configuration, (previous essay said every day update blog did not do but I really have in busy = =) The whole process out of countless, fortunately lucky enough configuration finished, while running cifar-10 train_ Full time to tidy up the whole process, and so the mentor's Titanx to be configured again.
ENV:ALIENWARE17-R3, Ubuntu16.04 64-bit, NVIDIA Geforce GTX 980m (Mentor research funds bought Titanx didn't arrive, my alien had to confiscate first, haha)
Configured with CUDA8.0, cuDNN5.0, opencv2.4.13
Reference to some of the intranet and external network information, I will try to write all myself, really encountered the need to refer to the content will be posted on the site.
First you have to have some basic Linux foundation, even if you play on MacOS Terminal also can, after all, is unix-based.
A little mention of the process of installing ubuntu16.04 yourself: Go to the next image file on Ubuntu website, make your own system disk with U-disk, and then follow the dress code, see http://www.linuxidc.com/Linux/2016-04/130520.htm for details.
First open terminal, enter the following dependencies:
1 sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler2 sudo apt-get install--no-install-recommends libboost-all-dev3 sudo apt-get install Libopenblas-dev Liblapack-dev Libatlas-base-dev4 sudo apt-get install Libgflags-dev libgoogle-glog-dev Liblmdb-dev
I. Nvidia Driver
After installing the Nvidia driver for Ubuntu, the Ubuntu default driver recognizes that Intel's graphics card must not work, please visit Nvidia official website http://www.nvidia.com/Download/index.aspx?lang= En-US query what you need to drive, now January 30, 2017 basically standard is 375, but still check the more secure
Installation method There are many statements on the Internet, there is said to blacklist this one, to close XSERVER,LIGHTDM and the like. I've spent two days here, and all of these methods have caused problems. Finally, the way I installed my success was
1. Runsudo apt-get purge nvidia-*
2. Run and then sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
3. Runsudo apt-get install nvidia-375
4. Run sudo reboot
Here is a very tricky point, my time is mainly spent here, looked up a lot of information, before all kinds of machine reasons I only understand a few of them, here a little tidy up:
1. Choose Nvidia driver with additional drivers set in Ubuntu, fail to install completely
2. Blacklist the Nouveau, use Alt+ctrl+f1,sudo service LIGHTDM stop (Close xserver), apt-get manual download of the installation package, the same as when the machine
3. Install cuda8.0 when the first option "Install Nvidia driver" selected Yes ... Be sure to choose No here. Although his default is yes, he later discovered that the Caffe website has made it clear that the Nvidia driver version of Cuda binding is usually out-of-date so skip this step.
Can only say that if there is a special situation, please leave a message, I will try to help you solve. Blog I look at every day, reply no more than 24 hours
There are two methods to check whether the installation is complete, the first is the input sudo nvidia-smi, if the correct display of video card information is installed successfully
Second, directly in the upper right corner to view this computer, the video card is not the N card on the line
II. cuda8.0
When your Ubuntu can correctly recognize the Nvidia graphics card, we can start to install Cuda (in fact, there is a random order of information, but anyway I was in this order to configure the success, so to write).
First go to cuda official website download cuda8.0 https://developer.nvidia.com/cuda-downloads
Step by Step Select Linux->x86_64->ubuntu->16.04->runfile (here to choose Runfile to be able to skip Cuda's own n-card driver installation), then you can download, 1.4 g, You can drink a cup of tea now.
Download good after CD to your download place, run "sudo sh cuda_8.0.44_linux.run", the specific file name check, at the beginning there will be a lot to read the Overlord clauses, direct q to the last, input accept. The following will let you choose a variety of details of the installation items, here Note whether you want to install NVIDIA driver be sure to choose No, I installed the first.
After installing sudo gedit ~/.BASHRC, write the following to BASHRC last
Export Path=/usr/local/cuda-8.0/bin${path:+:${path}}
Export Ld_library_path=/usr/local/cuda8.0/lib64${ld_library_path:+:${ld_library_path}}
Then test the Cuda-given sample:
Cd/usr/local/cuda-8.0/samples/1_utilities/devicequery
Make
sudo./devicequery
If your GPU-related information is displayed, the installation is successful
III. CuDNN
Here I cudnn using the 5.0 version for CUDA8.0., in Https://developer.nvidia.com/rdp/cudnn-download download, you need to register an account first. Select Cudnn v5 Library for Linux.
Download and unzip sudo tar-zxvf later./File name
And then
$ cd folder/extracted/contents
$ sudo cp -P include/cudnn.h /usr/include
$ sudo cp -P lib64/libcudnn* /usr/lib/x86_64-linux-gnu/
$ sudo chmod a+r /usr/lib/x86_64-linux-gnu/libcudnn*
Iv. OpenCV
The version selected here is 2.4.13. There are some compatibility issues with 3.0,3.1, so I chose the 2016 release of 2.4.13
http://opencv.org/downloads.html Download here
Download Good later
Unzip Opencv-2.4.13.zip specific file name see for yourself, there will be discrepancies
Copy it to the ready to install directory, where the assume folder is named OpenCV
CD ~/OPENCV
mkdir Build
CD Build
sudo apt install cmake
sudo cmake-d cmake_build_type=release-d
Cmake_install_prefix=/usr/local.
sudo make
My 2.4 version of the direct configuration, if it is 3.0 and 3.1 may appear compatibility issues, specific to the individual choose to consult the data to solve, here I did not install 3.0
After compile
sudo make install
Complete the installation
V. caffe!
Foreplay so much finally getting to the point of starting to configure Caffe.
Download Caffe:git clone https://github.com/BVLC/caffe.git from GitHub
If you don't have Git installed, you'll need to: sudo apt-get install git
Then sudo cp Makefile.config.example makefile.config
Open the Makefile.config file, here you need permission do not forget sudo. You can use Getit or nano.
If you have not skipped cudnn before, you will need to
#USE_CUDNN: = 1 # removed
If your OPENCV version is 3, you need to remove the opencv_version:=3.
The other configuration is self-viewing, according to the previous steps only need to cudnn that line of comments removed can
And then
Will # Whatever else you find need goes here.
Include_dirs: = $ (python_include)/usr/local/include
Library_dirs: = $ (python_lib)/usr/local/lib/usr/lib
Modified to:
Include_dirs: = $ (python_include)/usr/local/include/usr/include/hdf5/serial
Library_dirs: = $ (python_lib)/usr/local/lib/usr/lib/usr/lib/x86_64-linux-gnu/usr/lib/x86_64-linux-gnu/hdf5/ Serial
And then
(5) Modify the makefile file
Open the Makefile file and make the following changes:
Will:
Nvccflags +=-ccbin=$ (CXX)-xcompiler-fpic $ (common_flags)
To be replaced by:
Nvccflags + =-d_force_inlines-ccbin=$ (CXX)-xcompiler-fpic $ (common_flags)
(6) Edit/usr/local/cuda/include/host_config.h
Comment out the 115th line:
Will
#error--Unsupported GNU version! GCC versions later than 4.9 is not supported!
Switch
#error--Unsupported GNU version! GCC versions later than 4.9 is not supported!
Last sudo make all
The problem I encountered:
"libcudart.so.8.0 cannot open shared object File:no such file or directory"
The workaround is to copy some files to the/usr/local/lib folder:
#注意自己CUDA的版本号!
sudo cp/usr/local/cuda-8.0/lib64/libcudart.so.8.0/usr/local/lib/libcudart.so.8.0 && sudo ldconfig
sudo cp/usr/local/cuda-8.0/lib64/libcublas.so.8.0/usr/local/lib/libcublas.so.8.0 && sudo ldconfig
sudo cp/usr/local/cuda-8.0/lib64/libcurand.so.8.0/usr/local/lib/libcurand.so.8.0 && sudo ldconfig
Test after completion:
sudo make runtest
After running, if the bottom left corner shows the green passed, then the Caffe configuration is successful and done!
VI. Small trial Sledgehammer
Here Caffe has been configured, let's find a set of datasets to test, here choose Cifar-10.
Specific Process Caffe Official website has http://caffe.berkeleyvision.org/gathered/examples/cifar10.html
I'll simply write
First download install cifar-10, here I assume you install Caffe folder is called Caffe
CD Caffe
./data/cifar10/get_cifar10.sh
./examles/cifar10/create_cifar10.sh
Then it's OK, easy, come up with a quick set of tests:
CD Caffe
./examples/cifar10/train_quick.sh
A cup of tea he should have run out, set the iteration only 5000, the last will show optimization done.
Vii. Summary
No more said, pit giant, specific questions please leave a message I try to help
Reference
Http://www.2cto.com/kf/201610/552429.html
http://askubuntu.com/questions/767269/how-can-i-install-cudnn-on-ubuntu-16-04
Caffe configuration process in ubuntu16.04