Install Openface on Ubuntu 16.04

Introduction

My hardware platform is Intel Xeon E5-2618L 10-core processor and NVidia GTX1060 6GB graphic card. The OS distro is Ubuntu Mate 16.04.1 LTS x64.

The GTX1060 driver, Ubuntu system, GCC 5.4, torch and CUDA SDK have a compatibility problem that blocks me to use GPU acceleration for Openface.

To use GTX1060 with CUDA, you need to install CUDA8.0 SDK. Visit NVidia web site for more information:

Preparation

Install dependence:

sudo apt-get update
sudo apt-get upgrade
sudo apt-get install build-essential cmake curl gfortran git libatlas-dev libavcodec-dev libavformat-dev libboost-all-dev libgtk2.0-dev libjpeg-dev liblapack-dev libswscale-dev pkg-config python-dev python-pip wget -y

Install appropriate NVidia drivers and tools for GTX1060:

sudo apt-get purge nvidia*
sudo apt-get install nvidia-367 nvidia-367-dev nvidia-opencl-icd-367 nvidia-settings
sudo apt-get install

Install python packages:

pip install numpy scipy pandas scikit-learn scikit-image

Install OpenCV

Install dependence, some of them may be installed in previous step:

# Required:
sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev
# Optional
sudo apt-get install python-dev python-numpy libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libjasper-dev libdc1394-22-dev

Fetch OpenCV and switch to version 2.4.13:

git clone https://github.com/opencv/opencv.git
cd opencv
git checkout 2.4.13

Use cmake to configure OpenCV with CUDA/OpenCL or OpenMP (multi CPU cores) support.

mkdir build
cd build
cmake -D WITH_CUDA=1 -D WITH_OPENMP=1 -D WITH_OPENCL=1 -D CMAKE_BUILD_TYPE=RELEASE -D CMAKE_INSTALL_PREFIX=/usr/local ..

Compile and install OpenCV, and set -jxx according to your host PC's multi-thread capability:

make -j20
sudo make install

For more information refer to the OpenCV installation guide.

Install dlib

Fetch dlib and switch to version 18.18:

git clone https://github.com/davisking/dlib.git
cd dlib
git checkout v19.0

Configure, compile and install dlib:

cd python_examples
cmake ../tools/python  
cmake --build . --config Release -- -j20  
cp dlib.so /usr/local/lib/python2.7/dist-packages

Unfortunately, dlib doesn't support GPU acceleration at all.

Install OpenCL/CUDA enabled Torch distro:

Fetch cltorch and install dependens:

git clone --recursive https://github.com/hughperkins/distro -b distro-cl ~/torch-cl
cd ~/torch-cl
bash install-deps

Because there's compatible issue between GCC4.9, boost and dlib, we should change the default gcc to GCC5.4 before installing cltorch. Edit ./install.sh and replace gcc-4.9 & g++-4.9 to gcc & g++.

if [[ $(gcc -dumpversion | cut -d . -f 1) == 5 ]]; then {
#  export CC=gcc-4.9
#  export CXX=g++-4.9
  export CC=gcc
  export CXX=g++
} fi
...
if [[ $(gcc -dumpversion | cut -d . -f 1) == 5 ]]; then {
#  echo export CC=gcc-4.9>>$PREFIX/bin/torch-activate
#  echo export CXX=g++-4.9>>$PREFIX/bin/torch-activate
  echo export CC=gcc>>$PREFIX/bin/torch-activate
  echo export CXX=g++>>$PREFIX/bin/torch-activate
} fi

Then install cltorch:

./install.sh

You should also link the executable files which are needed by openface:

ln -s ~/torch-cl/install/bin/* /usr/local/bin

To verify OpenCL features, use these commands below:

source ~/torch-cl/install/bin/torch-activate
luajit -l torch -e 'torch.test()'
luajit -l nn -e 'nn.test()'
luajit -l cltorch -e 'cltorch.test()'
luajit -l clnn -e 'clnn.test()'

The cutorch & cunn are also available:

luajit -l cutorch -e 'cutorch.test()'
luajit -l cunn -e 'nn.testcuda()'

To update cltorch use the command:

cd ~/torch-cl
git pull
git submodule update --init --recursive
./install.sh