Skip to content

Latest commit

 

History

History
81 lines (74 loc) · 2.84 KB

INSTALL.md

File metadata and controls

81 lines (74 loc) · 2.84 KB
Raw

Installing Caffe, bilateralNN, and their dependencies

Step-by-step installation instructions for Ubuntu 16.04.

  • Make sure CUDA and cuDNN are installed

    We tested on CUDA8/cuDNN6 and CUDA9/cuDNN7. But Caffe typically supports a wide range of CUDA/cuDNN versions.

  • Install system-wide dependencies

    sudo apt-get update
sudo apt-get install -y --no-install-recommends \
    build-essential \
    cmake \
    git \
    ca-certificates \
    wget \
    rsync \
    vim \
    libgl1-mesa-glx \
    libgtk2.0-0 \
    libatlas-base-dev \
    libboost-all-dev \
    libgflags-dev \
    libgoogle-glog-dev \
    libleveldb-dev \
    liblmdb-dev \
    libprotobuf-dev \
    libsnappy-dev \
    libpng12-dev \
    protobuf-compiler

  • Install conda and additional packages

    wget --quiet https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
chmod a+x miniconda.sh
./miniconda.sh  # you will be prompted to agree on terms and to configure the installation
rm miniconda.sh
source ~/.bashrc
conda update --yes conda
conda create -n caffe -y python=3.5
conda install -n caffe -c menpo -y opencv3
conda install -n caffe -y cython scikit-image scikit-learn matplotlib bokeh ipython h5py nose pandas pyyaml jupyter

  • Install Caffe and bilateralNN

    CAFFE_ROOT=~/caffe  # specify where caffe will be installed (does not need to be inside the main project)
git clone -b 1.0 --depth 1 https://github.com/BVLC/caffe.git $CAFFE_ROOT
cd $CAFFE_ROOT
git clone --depth 1 https://github.com/suhangpro/caffe.git caffe-patch
rsync -av caffe-patch/bpcn/ ./
rm -rf caffe-patch
source activate caffe
mkdir build && cd build
cmake -DUSE_CUDNN=1 -Dpython_version=3 ..
make -j$(nproc)
cd $CAFFE_ROOT/python
sed -i -e "s/python-dateutil>=1.4,<2/python-dateutil>=2.0/g" requirements.txt
for req in $(cat requirements.txt); do pip install $req; done  # 'msgpack' warning message can be safely ignored
cd $CAFFE_ROOT
echo "
# caffe paths
export PYCAFFE_ROOT=$(pwd)/python
export PYTHONPATH=\$PYCAFFE_ROOT:\$PYTHONPATH
export PATH=\$CAFFE_ROOT/build/tools:\$PYCAFFE_ROOT:\$PATH
" >> ~/.bashrc  # this puts path in your .bashrc file
source ~/.bashrc
sudo /bin/bash -c 'echo "$(pwd)/build/lib" >> /etc/ld.so.conf.d/caffe.conf'
sudo ldconfig

  • Ready to use!

    # enter conda environment if you are not already in one
source activate caffe

Note that for multi-GPU usage (not used or tested in our experiments and demos) you will also need NCCL:

  • Before installing Caffe, install NCCL following official instructions.
  • Add -DUSE_NCCL=1 flag for cmake.