installation.md

Installation of Minkowski Engine

We use Minkowski Engine for sparse convolution of point cloud in our project.

MinkowskiEngine==0.5.4 with cudatoolkit=10.2 was used for the project.

First we creat a new environment

conda create -n box2mask python=3.7
conda activate box2mask

Setup the CUDA system environment variables like the example below:

cuda_version=10.2
# please set the right path to CUDA in your system, bellow is an example used for our system
export CUDA_HOME=/usr/lib/cuda-${cuda_version}/ 
export PATH=/usr/lib/cuda-${cuda_version}/bin/:${PATH}
export LD_LIBRARY_PATH=/usr/lib/cuda-${cuda_version}/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
export CUDA_PATH=/usr/lib/cuda-${cuda_version}/

Next, we install pytorch with cudatoolkit and dependencies

conda install pytorch=1.8.1 torchvision cudatoolkit=${cuda_version} -c pytorch -c nvidia

Install dependencies for Minkowski Engine

pip install torch ninja
conda install openblas-devel -c anaconda 

We then install gcc version 7 sudo apt install g++-7 # For CUDA 10.2, must use GCC <= 8

Make sure g++-7 --version is at least 7.4.0 export CXX=g++-7

Install Minkowski Engine via pip:

pip install -U MinkowskiEngine==0.5.4 --install-option="--blas=openblas" -v --no-deps

For more detailed installation instruction, see MinkowskiEngine.

Checking installations of Minkowski Engine

The following commands will clone the repository of Minkowski Engine and run an example segmentation model on an indoor point cloud:

git clone https://github.com/NVIDIA/MinkowskiEngine.git
cd MinkowskiEngine
# code requires open3d
pip install open3d
python -m examples.indoor

Install GIT repository and other dependencies

The following commands will clone Box2Mask repo on your machine and install the remaining dependencies. Note that you should still be using box2mask environemnt

git clone -b release https://github.com/jchibane/Box2Mask.git box2mask
cd box2mask
conda env update --file env.yml