Copyright statement: This article is the original article of the blogger and may not be reproduced without the permission of the blogger. https://blog.csdn.net/Gavin__Zhou/article/details/47363385
In the past few days, I saw something about Deep Learning and I also wanted to have fun. I chose Caffe from Theano, Torch7 and Caffe. The main reason is that the installation is relatively simple and there are many online tutorials.
Post an introduction to Caffe:
Caffe is a clear and efficient deep learning framework. Its author is Jia Yangqing who graduated from UC Berkeley with a Ph.D. and currently works at Google.
Caffe is a pure C++/CUDA architecture, supporting command line, Python and MATLAB interfaces; it can be seamlessly switched between CPU and GPU:
Advantages of Caffe
Quick start: The model and the corresponding optimization are given in the form of text instead of code.
Caffe gives the definition of the model, optimal settings and pre-training weights, which is convenient to get started immediately.
Fast speed: able to run the best model and massive data.
Caffe and cuDNN are used in combination to test the AlexNet model. It only takes 1 per image to process on K40.17ms.
Modularity: Easy to expand to new tasks and settings.
You can use the various layer types provided by Caffe to define your own model.
Openness: Open codes and reference models are used for reproduction.
Community is good: you can pass BSD-2 Participate in development and discussion.
So let's go to the Internet to find a bunch of them, not our focus this time, skip it. . . . .
I also found a bunch of tutorials on the Internet. There are simple and complex ones. Many of them are outdated and cannot be used. I also refer to many online blog materials. Please contact me if there is any infringement.
Let me talk about the configuration of my own machine, i5+Intel set display+4GB RAM, so GPU acceleration like CUDA is not available, but it is not a problem for just getting started. Caffe recommends installing it under Linux such as Ubuntu or CentOS. The official tutorials are also the same. It is really not recommended under Windows.
Caffe official websiteCaffe official guide
Start the installation process below:
I won’t talk about it. If it doesn’t work, you can refer to this Ubuntu Installation
Although I don’t need my broken machine, I still install it, so that I don’t have to worry about any further business trips.
First, you need to download the CUDA installation package: CUDA download address, enter and select your own Ubuntu14.04 deb package to download.
Second, install the repository and CUDA:
$ sudo dpkg -i cuda-repo-ubuntu1410-7-0-local_7.0-28_amd64.deb
$ sudo apt-get update
$ sudo apt-get install -y cuda
Set environment variables:
$ export CUDA_HOME=/usr/local/cuda7
$ export LD_LIBRARY_PATH=${CUDA_HOME}/lib64
$ PATH=${CUDA_HOME}/bin:${PATH}
$ export PATH
The python system already comes with it. Python2.7 requires the installation of the ipython environment. Many Caffe programs require ipython. Someone on the Internet directly installs spyder. Once installed, it will automatically install ipython for you by default, or you can install it. EPD Download the script and install it. By default, it will help you install many libraries, including ipython, numpy, scipy, pandas, matplotlib, etc.; if it is troublesome, you can also search for spyder installation in the ubuntu software center. can.
BLAS chooses MKL, you can also choose ATLAS or OpenBLAS, MKL download address, you need to apply first before downloading, be sure to keep the installation serial number, which will be used later
$ tar zxvf parallel_studio_xe_2015.tar.gz
$ chmod a+x parallel_studio_xe_2015 -R
$ sudo ./install_GUI.sh
In this way, you start the GUI installation, and it feels like the installation and software under windows, next all the way, and finally enter the serial number sent to you by the email before, and it will be OK.
Create a new MKL configuration file:
$ sudo gedit /etc/ld.so.conf.d/intel_mkl.conf
Write the following two sentences:
/opt/intel/lib/intel64
/opt/intel/mkl/lib/intel64
Create a new cuda configuration file:
$ sudo gedit /etc/ld.so.conf.d/cuda.conf
Write the following two sentences:
/usr/local/cuda/lib64
/lib
Finally complete the link:
$ sudo ldconfig -v
Here we use the installation package provided by others to solve some dependency problems. Download link: OpenCV
Switch to the folder where the file is saved, and install the dependencies:
$ sudo sh Ubuntu/dependencies.sh
Change directory 3.0\ install OpenCV 3.0.0rc1:
$ sudo sh opencv3_0_0-rc1.sh
Network installation, just wait
glog download address
installation:
$ tar zxvf glog-0.3.3.tar.gz
$ ./configure
$ make
$ sudo make install
Other dependencies:
$ sudo apt-get install -y libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libboost-all-dev libhdf5-serial-dev
$ sudo apt-get install -y libgflags-dev libgoogle-glog-dev liblmdb-dev protobuf-compiler
$ sudo apt-get install -y python-numpy python-scipy python-matplotlib python-sklearn python-skimage python-h5py python-protobuf python-leveldb python-networkx python-nose python-pandas python-gflags Cython ipython
$ sudo apt-get install -y protobuf-c-compiler protobuf-compiler
If you don’t need GPU acceleration in this step, you don’t need to install it. You can also run the CPU.
a. This modified version of caffe-master supports cudnn-6.5-linux-x64-v2 by default. If you use cudnn-6.5-linux-R1, an error will be reported. Please go to the official website to download the latest cuDNN before installation.
$ sudo cp cudnn.h /usr/local/include
$ sudo cp libcudnn.so /usr/local/lib
$ sudo cp libcudnn.so.6.5/usr/local/lib
$ sudo cp libcudnn.so.6.5.48/usr/local/lib
b. Link cuDNN library file
$ sudo ln -sf /usr/local/lib/libcudnn.so.6.5.48/usr/local/lib/libcudnn.so.6.5
$ sudo ln -sf /usr/local/lib/libcudnn.so.6.5/usr/local/lib/libcudnn.so
$ sudo ldconfig -v
c. Switch to the Caffe-master folder, generate the Makefile.config configuration file, and execute:
$ cp Makefile.config.example Makefile.config
d. Configure the Makefile.config file (only the modified part is listed)
<1>. Enable CUDNN, remove "#" (currently caffe-master still only supports R1 version)
USE_CUDNN :=1
<2>. Enable GPU, add comment "#"
# CPU_ONLY :=1
<3>. Configure some reference files (the added part is mainly to solve the path problem of HDF5 under the new version)
INCLUDE_DIRS :=$(PYTHON_INCLUDE)/usr/local/include /usr/lib/x86_64-linux-gnu/hdf5/serial/include
LIBRARY_DIRS :=$(PYTHON_LIB)/usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial
<4>. Enable Intel Parallel Studio XE 2015 Professional Edition for C++ Linux
BLAS := mkl
<5>. Configure the path to realize caffe's support for Python and Matlab interfaces
PYTHON_LIB :=/usr/local/lib
MATLAB_DIR :=/usr/local/MATLAB/R2014a
<6>. Configure Makefile (to achieve support for OpenCV 3.x)
Find the section "Derive include and lib directories", modify the last line of "LIBRARIES +=", add opencv_imgcodecs
opencv_core opencv_highgui opencv_imgproc opencv_imgcodecs
caffe source code download address
Compile:
$ make all -j8
$ make test -j8
$ make runtest -j8
$ make pycaffe -j8
$ make matcaffe -j8
This should be fine, and I will write about how to use Caffe to run my own things in the next blog.
Reference：
(1) [OuXinYU's blog] (https://ouxinyu.github.io/Blogs/20140723001.html)
(2) [CSDN's Caffe introduction] (http://www.csdn.net/article/2015-01-22/2823663)
(3) [Other] (http://blog.csdn.net/kuaitoukid/article/details/40395617)
Recommended Posts