Install Caffe - Ubuntu-16.04 - Python 3.5

This is a guide to install Caffe with GPU support on Ubuntu 16.04 with Python 3.5. Change the commands accordingly to use this guide for other version of Python 3.

Video Tutorial: https://www.youtube.com/watch?v=ytyU9NLkHl0

This guide assumes that you have already installed CUDA and cuDNN.

For CUDA installation follow: http://docs.nvidia.com/cuda/cuda-installation-guide-linux/#axzz4VZnqTJ2A.

For cuDNN follow: http://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html.

Step 1: Update and Upgrade the packages list.

sudo apt-get update
sudo apt-get upgrade

Step 2: Install general dependencies. Open terminal and type following commands in terminal.

sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev protobuf-compiler
sudo apt-get install --no-install-recommends libboost-all-dev
sudo apt-get install libatlas-base-dev
sudo apt-get install libopenblas-dev
sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev

Step 3: Go to the installation folder using "cd /your/path" command. Download/Clone the latest git repository using following command and go to

the caffe directory in terminal.

git clone https://github.com/BVLC/caffe.git
cd caffe

Step 4: Install the python requirements for caffe. Use following commands in terminal.

cd python
for req in $(cat requirements.txt); do pip3 install --no-cache-dir $req; done (Wait for it to finish installation)
cd ..

Step 5: Add caffe python path to the $PYTHONPATH(or Step 11)

export PYTHONPATH=/path/to/caffe/python

Example: export PYTHONPATH=/home/user/xyz/caffe/python

Step 6: Get the Python 3.5 USER_SITE path by typing following command in the terminal.

python3 -m site

We will need the path given in USER_SITE. e.g. USER_SITE: '/home/user/.local/lib/python3.5/site-packages'. If USER_SITE doesn't exist, then use the one with /dist-packages in the end.

Step 7: Generate the Makefile configuration file using following command in terminal.

cp Makefile.config.example Makefile.config

Step 8: We will adjust the Makefile.config file in order to install the caffe for Python 3. Edit the Makefile.config in your favourite editor.

Step 9: Perform the following changes in Makefile.config.

1. Comment the PYTHON_INCLUDE lines that refer to python2.7. They should look like this:

#PYTHON_INCLUDE := /usr/include/python2.7
#/usr/lib/python2.7/dist-packages/numpy/core/include

2. Uncomment PYTHON_LIBRARIES and PYTHON_INCLUDE for python 3. They should look like this at this point:

PYTHON_LIBRARIES := boost_python3 python3.5m

PYTHON_INCLUDE := /usr/include/python3.5m
/usr/lib/python3.5/dist-packages/numpy/core/include

3. Use the path copied in step 6 and add following path to the PYTHON_INCLUDE.

/home/user/.local/lib/python3.5/site-packages/numpy/core/include

PYTHON_INCLUDE should look like this at this point.

PYTHON_INCLUDE := /usr/include/python3.5m
/usr/lib/python3.5/dist-packages/numpy/core/include
/home/nvme/.local/lib/python3.5/site-packages/numpy/core/include

4. Uncomment the line WITH_PYTHON_LAYER := 1.

5. Add '/usr/include/hdf5/serial' to the INCLUDE_DIRS variable. It should look like this:

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial

6. Find the path to hdf5, use command: "find /usr/lib -name hdf5". Copy the path and add it to LIBRARY_DIRS after appending '/serial' after it. It should look like this at this point.

LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial

7. Find the path to libpython3.5m by using command "find /usr -name libpython3.5*". It will return multiple paths to the library but use '/usr/lib/x86_64-linux-gnu/'. Add this path to PYTHON_LIB and LIBRARY_DIRS. Both of these should look like this at this point.

PYTHON_LIB := /usr/lib /usr/lib/x86_64-linux-gnu/

LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial /usr/lib/x86_64-linux-gnu/

8. Find the path to libboost_python-py35 using command "find /usr -name libboost_python-py35*". Usually the path is '/usr/lib/x86_64-linux-gnu/' which have already been added to the PYTHON_LIB and LIBRARY_DIRS in 7. If it is different then add it to both of these variables.

9. Now rename boost_python3 in PYTHON_LIBRARIES to boost_python-py35. PYTHON_LIBRARIES should look like this.

PYTHON_LIBRARIES := boost_python-py35 python3.5m

10. Make sure CPU_ONLY := 1 is commented in order to install caffe with GPU support.

Step 10: Install caffe using following commands. Make sure you are in caffe home directory.

make all
make runtest

Step 11: Install caffe

By default, Caffe will not be installed (no target 'install' provided). We can copy the .so file manually

sudo cp build/lib/libcaffe.so* /usr/lib

Step 12: Install pycaffe

1. Compile:

make pycaffe

2. install: Install pycaffe manually by copying to dist-packages

sudo cp -r python/caffe/ /usr/local/lib/python3.5/dist-packages/

Step 13: Check by importing caffe in python3 interpreter. Open python3 interpreter and import caffe.

1. python3
2. import caffe

or python3 -c "import caffe" At this point caffe should be imported and be able to use successfully. Sample "Makefile.config" is added for reference.