This document describes how to install Video-LFB and its dependencies.
Requirements:
- NVIDIA GPU, Linux, Python2
- Caffe2, various standard Python packages.
- We used CUDA and cuDNN in our experiments.
Clone the Video-LFB repository.
# VIDEO_LFB=/path/to/install/video-long-term-feature-banks
git clone https://github.com/facebookresearch/video-long-term-feature-banks $VIDEO_LFB
Add this repository to $PYTHONPATH.
export PYTHONPATH=/path/to/video-long-term-feature-banks/lib:$PYTHONPATHCaffe2 is now part of PyTorch.
Please follow PyTorch official instructions to install from source.
The only additional step is to add a customized operator (affine_nd_op) to the source code after cloning the source:
git clone --recursive https://github.com/pytorch/pytorch
rm -rf pytorch/caffe2/video
cp -r [path to video-long-term-feature-banks]/caffe2_customized_ops/video pytorch/caffe2/In case it's still not clear, in the following we provide the exact steps we performed to install Caffe2. We used an Ubuntu machine with CUDA 9.0 and cuDNN 7.5, without root permission.
conda create -n video-lfb python=2.7
source activate video-lfb
conda install numpy pyyaml mkl mkl-include setuptools cmake cffi typing
conda install -c pytorch magma-cuda90
git clone --recursive https://github.com/pytorch/pytorch
cd pytorch
rm -r caffe2/videos
cp -r [path to video-long-term-feature-banks]/caffe2_customized_ops/video caffe2/
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
export CUDNN_LIB_DIR=[path to cnDNN]/lib64
export CUDNN_INCLUDE_DIR=[path to cnDNN]/include
python2 setup.py install
conda install --yes protobuf
conda install --yes future
conda install --yes networkx
pip install enum34
pip install sklearn