Skip to content

Latest commit

 

History

History
71 lines (64 loc) · 4.21 KB

INSTALL.md

File metadata and controls

71 lines (64 loc) · 4.21 KB
Raw

Installation

Basic dependencies

We recommend creating a new conda environment for a clean installation of the dependencies. All following commands are assumed to be executed within this conda environment. The code has been tested on CentOS 7.8, python 3.7 and CUDA 10.2. We used RTX 2080 Ti for training and testing our models.

conda create --name nba python=3.7
conda activate nba

Make sure CUDA 10.2 is your default cuda. If your CUDA 10.2 is installed in /usr/local/cuda-10.2, add the following lines to your ~/.bashrc and run source ~/.bashrc:

export PATH=$PATH:/usr/local/cuda-10.2/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.2/lib64
export CPATH=$CPATH:/usr/local/cuda-10.2/include

Install dependencies, we use ${REPO_ROOT_DIR} to represent the working directory of this repo.

cd ${REPO_ROOT_DIR}
conda install pytorch==1.5.1 torchvision==0.6.1 cudatoolkit=10.2 -c pytorch
pip install -r requirements.txt

Mesh penetration optimization dependencies

Please install our modified torch-mesh-isect CUDA extension for pytorch 1.5.1. Before installing the extension, make sure to set the environment variable $CUDA_SAMPLES_INC to the path that contains the header helper_math.h, which can be found in the repo CUDA Samples repository. If your helper_math.h is located at /usr/local/cuda-10.2/samples/common/inc, add the following lines to your ~/.bashrc and run source ~/.bashrc:

export CUDA_SAMPLES_INC=/usr/local/cuda-10.2/samples/common/inc

After that, run the following commands:

cd img_to_mesh/extensions/torch-mesh-isect
python setup.py install

Mesh evaluation dependencies

Please install open3d 0.9.0 for ICP. You also need to install the Chamfer Distance and Earth-Mover Distance CUDA extension:

pip install https://github.com/intel-isl/Open3D/releases/download/v0.9.0/open3d-0.9.0.0-cp37-cp37m-manylinux1_x86_64.whl
cd img_to_mesh/extensions/chamfer
python setup.py install
cd img_to_mesh/extensions/emd
python setup.py install

If your gcc version is 9, you may encounter errors when you build the extensions for cuda 10.2 and pytorch 1.5.1. To solve this problem, you can install gcc 8 on your machine using following commands:

sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 8
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-8 8

then select gcc version using the following command:

sudo update-alternatives --config gcc

Demo dependencies

Please make sure OpenPose is installed on your system. For our case, we use the Windows binary and run the preprocessing code on a Windows machine. If you have difficulties installing OpenPose, you can also try Alphapose and make sure the output json file is in openpose format. You may need to modify our preprocessing code a little bit if you are using Alphapose (We do not try with Alphapose though).

Customized training dependencies

If you want to customize the hyperparameters of the spiral convolution training, please make sure the following packages are installed in python 3.

  • chumpy: This link provides some solution for installing chumpy when pip>=20.1.
  • opendr: This link provides some solution for installing opendr in python 3.
  • PSBody Mesh package

Download pretrained checkpoints

We provide pretrained checkpoints at the Google Drive. After you download the checkpoints. please copy the folder to the target paths:

cd ${REPO_ROOT_DIR}
mkdir -p img_to_mesh/log
cp -r pretrained_ckpt/pose img_to_mesh/log
cp -r pretrained_ckpt/mesh img_to_mesh/log
mkdir -p global_position/field_lines/log
cp -r pretrained_ckpt/court_lines global_position/field_lines/log