Ubuntu16.04+cuda8.0rc+opencv3.1.0+caffe+Theano+torch7搭建教程

$ sudo add-apt-repository ppa:graphics-drivers/ppa

$ sudo apt-get update

$ nvidia-settings

# service lightdm stop

# vim /etc/modprobe.d/blacklist.conf

blacklist nouveau

1.  
   # chmod +x cuda*.run #获取文件权限
2.  
   # ./cuda*.run #执行文件安装

# vim /etc/profile

1.  
   PATH=/usr/local/cuda/bin:$PATH
2.  
   export PATH

# source /etc/profile

# vim /etc/ld.so.conf.d/cuda.conf

 /usr/local/cuda/lib64

# ldconfig

# cd /usr/local/cuda/samples

# make all -j2   (我的本本是双核CPU，所以-j2都用上加快编译速度)

# ./deviceQuery

1.  
   $ sudo apt-get -y remove ffmpeg x264 libx264-dev
2.  
   $ sudo apt-get -y install libopencv-dev build-essential checkinstall cmake pkg-config yasm libtiff4-dev libjpeg-dev libjasper-dev libavcodec-dev libavformat-dev libswscale-dev libdc1394-22-dev libxine-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev libv4l-dev x264 v4l-utils ffmpeg libgtk2.0-dev

1.  
   $ mkdir build
2.  
   $ cd build
3.  
   $ cmake -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local -D WITH_TBB=ON -D BUILD_NEW_PYTHON_SUPPORT=ON -D WITH_V4L=ON -D INSTALL_C_EXAMPLES=ON -D INSTALL_PYTHON_EXAMPLES=ON -D BUILD_EXAMPLES=ON -D WITH_QT=ON -D WITH_OPENGL=ON ..

1.  
   $ make -j2
2.  
   $ sudo make install

1.  
   $ sudo sh -c ‘echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf‘
2.  
   $ sudo ldconfig

1.  
   #include "precomp.hpp"
2.  
   #if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)

1.  
   #include "precomp.hpp"
2.  
   #if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000)

$ sudo apt-get install python-pip

1.  
   $ sudo pip install tornado pyzmq pygments
2.  
   $ sudo apt-get install libzmq-dev

$ sudo ipython notebook

$ sudo apt-get install python-numpy python-scipy python-matplotlib python-qt4 qt4-designer pyqt4-dev-tools python-qt4-doc spyder cython swig python-sklearn python-skimage python-h5py python-protobuf python-leveldb python-networkx python-nose python-pandas python-gflags ipython protobuf-c-compiler protobuf-compiler

1.  
   $ sudo apt-get update
2.  
   $ sudo apt-get install build-essential cmake git pkg-config libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler libgflags-dev libgoogle-glog-dev liblmdb-dev python-dev
3.  
   $ sudo apt-get install --no-install-recommends libboost-all-dev

$ sudo apt-get install libatlas-base-dev

$ cp Makefile.config.example Makefile.config

1.  
   <pre name="code" class="html">注释第5行的 USE_CUDNN := 1
2.  
   去掉注释第21行的 OPENCV_VERSION := 3

1.  
   在 INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include 后面添加 /usr/include /usr/include/hdf5/serial
2.  
   在 LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib 后面添加 /usr/lib/x86_64-linux-gnu/hdf5/serial
3.  
   实现caffe对Python和Matlab接口的支持 PYTHON_LIB := /usr/local/lib

1.  
   $ make all -j2
2.  
   $ make test -j2
3.  
   $ make runtest -j2

$ make pycaffe -j2

error -- unsupported GNU version! gcc versions later than 5.3 are not supported!

1.  
   #if __GNUC__ > 5 || (__GNUC__ == 5 && __GNUC_MINOR__ > 3)
2.  
   //#error -- unsupported GNU version! gcc versions later than 5.3 are not supported!
3.  
   #endif /* __GNUC__ > 5 || (__GNUC__ == 5 && __GNUC_MINOR__ > 1) */

$ sudo pip install theano

$ sudo vim ~/.theanorc 

1.  
   [global]
2.  
   openmp=False
3.  
   device=gpu
4.  
   floatX=float32
5.  
   allow_input-downcast=True
6.  
   [blas]
7.  
   ldflags=
8.  
   [nvcc]
9.  
   flags=-D_FORCE_INLINES

1.  
   from theano import function, config, shared, sandbox
2.  
   import theano.tensor as T
3.  
   import numpy
4.  
   import time
5.  
   vlen = 10 * 30 * 768 # 10 x #cores x # threads per core
6.  
   iters = 1000
7.  
   rng = numpy.random.RandomState(22)
8.  
   x = shared(numpy.asarray(rng.rand(vlen), config.floatX))
9.  
   f = function([], T.exp(x))
10.  
    print(f.maker.fgraph.toposort())
11.  
    t0 = time.time()
12.  
    for i in range(iters):
13.  
    r = f()
14.  
    t1 = time.time()
15.  
    print("Looping %d times took %f seconds" % (iters, t1 - t0))
16.  
    print("Result is %s" % (r,))
17.  
    if numpy.any([isinstance(x.op, T.Elemwise) for x in f.maker.fgraph.toposort()]):
18.  
    print(‘Used the cpu‘)
19.  
    else:
20.  
    print(‘Used the gpu‘)

$ python test.py

1.  
   /usr/bin/python2.7 /home/hjimce/PycharmProjects/untitled/.idea/temp.py
2.  
   Using gpu device 0: GeForce 840M (CNMeM is disabled, cuDNN not available)
3.  
   [GpuElemwise{exp,no_inplace}(<CudaNdarrayType(float32, vector)>), HostFromGpu(GpuElemwise{exp,no_inplace}.0)]
4.  
   Looping 1000 times took 0.302778 seconds
5.  
   Result is [ 1.23178029 1.61879349 1.52278066 ..., 2.20771813 2.29967761
6.  
   1.62323296]
7.  
   Used the gpu

1.  
   $ sudo apt-get install curl
2.  
   $ curl -sk https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash

1.  
   $ git clone https://github.com/torch/distro.git ~/torch --recursive
2.  
   $ cd ~/torch; ./install.sh

$ source ~/.bashrc

1.  
   $ th
2.  
   $ totch.Tensor{1,2,3}

1.  
   $ luarocks install image
2.  
   $ luarocks list

1.  
   $ luarocks install nngraph
2.  
   $ luarocks install optim