标签：
This Guide is based on caffe github wiki guide (https://github.com/BVLC/caffe/wiki/Ubuntu-16.04-or-15.10-Installation-Guide )
Some parts of it have been changed to suit my computer.
The following guide includes the how-to instructions for the installation of BVLC/Caffe in Ubuntu 16.04 (preliminary procedure does not function with the current Cuda Toolkit) or 15.10 Linux (works well). This also includes the KUbuntu 16.04 or 15.10 and the related distributions.
Execute these commands first:
sudo apt-get update
sudo apt-get upgrade
sudo apt-get install -y build-essential cmake git pkg-config
sudo apt-get install -y libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler
sudo apt-get install -y libatlas-base-dev
sudo apt-get install -y --no-install-recommends libboost-all-dev
sudo apt-get install -y libgflags-dev libgoogle-glog-dev liblmdb-dev
# (Python general)
sudo apt-get install -y python-pip
# (Python 2.7 development files)
sudo apt-get install -y python-dev
sudo apt-get install -y python-numpy python-scipy
I didn‘t try python 3.5 to avoid a version confilct problem, and I also didn‘t use opencv 3.1 for the same reason.
I just want to make it simple and function well.
# (OpenCV 2.4)
sudo apt-get install -y libopencv-dev
Go to the https://github.com/BVLC/caffe and download zip archive. Unpack it to ~/bin/ or any other location. Enter the caffe-master directory in the terminal window.
Copy the Makefile.config.example to Makefile.config like this:
cp Makefile.config.example Makefile.config
and open it for editing (with a text editor). I use the kate editor for this purpose, so the command that I execute goes as follows. You first need to install the kate editor with:
sudo apt-get install kate
and then you can edit the configuration file with:
kate ./Makefile.config &
The following line in the configuration file tells the program to use CPU only for the computations.
CPU_ONLY := 1
This is the typical setting for a computer without any NVIDIA graphics card and it is typical for the installation of Caffe inside the typical virtual machine. (Notice that there is a special type of virtual machine inside the Ubuntu host machine that can access the physical NVIDIA graphics card directly. See https://github.com/NVIDIA/nvidia-docker)
Change the line accordingly by commenting it out (# CPU_ONLY := 1) if you have an NVIDIA graphics card with the proprietary driver, CUDA toolkit and CUDNN installed. The Makefile.config should contain the following lines, so find them and fill them in.
PYTHON_INCLUDE := /usr/include/python2.7 /usr/lib/python2.7/dist-packages/numpy/core/include
(For ways to create an isolated Python environment, explore the topic of virtual environments here: http://docs.python-guide.org/en/latest/dev/virtualenvs/)
WITH_PYTHON_LAYER := 1
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
Now lets continue with the instructions for version 15.10 first, followed by instructions for 16.04 users.
Execute the additional commands:
find . -type f -exec sed -i -e ‘s^"hdf5.h"^"hdf5/serial/hdf5.h"^g‘ -e ‘s^"hdf5_hl.h"^"hdf5/serial/hdf5_hl.h"^g‘ ‘{}‘ \;
cd /usr/lib/x86_64-linux-gnu
sudo ln -s libhdf5_serial.so.8.0.2 libhdf5.so
sudo ln -s libhdf5_serial_hl.so.8.0.2 libhdf5_hl.so
The above commands will need to be executed for Ubuntu 16.04 as well, but the file versions for libhdf5_serial.so and libhdf5_serial_h1.so are different and so the last two lines will need to be altered. Visit /usr/lib/x86_64-linux-gnu/ and list the relevant contents of that directory using a command
such as ls | grep hdf5
The versions of libhdf5 that need to be linked to are 10.1.0 and 10.0.2 respectively.
So, for Ubuntu16.04, execute the additional commands like this:
find . -type f -exec sed -i -e ‘s^"hdf5.h"^"hdf5/serial/hdf5.h"^g‘ -e ‘s^"hdf5_hl.h"^"hdf5/serial/hdf5_hl.h"^g‘ ‘{}‘ \;
cd /usr/lib/x86_64-linux-gnu
such as ls | grep hdf5
sudo ln -s libhdf5_serial.so.10.1.0 libhdf5.so
sudo ln -s libhdf5_serial_hl.so.10.0.2 libhdf5_hl.so
Now for both platforms lets return to the unpacked Caffe directory caffe-master and enter these commands:
cd python
for req in $(cat requirements.txt); do pip install $req; done
NOTE: If the Ubuntu operating system was updated, perhaps the Python layer needs to be updated and recompiled, because the Python module no longer works. Perform this step again in that case.
for req in $(cat requirements.txt); do pip install $req; done
In case of any problems, try:
for req in $(cat requirements.txt); do sudo -H pip install $req --upgrade; done
The build process will fail in Ubuntu 16.05 due to the GCC 5.x compiler, when compiling Cuda 7.5 sources. The updated version of Cuda Toolkit 8.0RC is compatible with GCC 5.x compiler in Ubuntu 16.05. Once Cuda Toolkit 8.0RC is installed, Caffe will successfully build and run in Ubuntu 16.05.
In any case, the next step is to execute the following code to build Caffe:
cd ..
(now you are in caffe-master directory)
make all
make test
make runtest
make pycaffe -should be finished already, so you can omit this one
make distribute
Note that the build process can be sped up by appending -j $(($(nproc) + 1)) to the above commands, which distributes the build across the available processors on your system. For example:
make all
can become
make all -j $(($(nproc) + 1))
In order to make the Python work with Caffe, open the file ~/.bashrc for editing in your favorite text editor. There, add the following line at the end of file:
export PYTHONPATH=/path/to/caffe-master/python:$PYTHONPATH
You can also execute that same line immediately as a command for immediate effects.
caffe + ubuntu16.04 (version without GPU)
标签：
原文地址：http://www.cnblogs.com/balmy/p/5752589.html
