instruction.txt

#create a conda environment
conda create -n 3dAvatarPipeline python=3.7
#depending on the cuda version (check - nvcc --version)
conda install pytorch torchvision torchaudio cudatoolkit=10.1 -c pytorch

#Requirements for PIFuHD
Pillow                # PIL
scikit-image          #skimage
tqdm
opencv-python         # cv2

#For visualization
trimesh
PyOpenGL
ffmpeg
freeglut (use sudo apt-get install freeglut3-dev for ubuntu users)

#Download pre-trained model
sh ./scripts/download_trained_model.sh

#-------------------------------------------------------------------------------
#Requirements for IPNet
ipdb
chumpy
smplpytorch

cd IPNet
#Install kaolin
git clone --recursive https://github.com/NVIDIAGameWorks/kaolin
cd kaolin
git checkout v0.1 #checkout to v0.1
python setup.py develop
cd ..

#Install MPI mesh library
cd mesh
sudo apt-get install libboost-dev #install boost library
git clone https://github.com/MPI-IS/mesh.git
BOOST_INCLUDE_DIRS=/path/to/boost/include make all  #use command `whereis boost` to locate path, generally in /usr/include/boost
make tests #check if installed correctly, might need to reinstall PyOpenGL if you face any errors here
cd ..
cd ..

#Download SMPL models
Create account here and accept license and download SMPL models - https://smpl.is.tue.mpg.de/

#Download checkpoints
Download IPNet weights: https://datasets.d2.mpi-inf.mpg.de/IPNet2020/IPNet_p5000_01_exp_id01.zip
or IPNet single surface: https://nextcloud.mpi-klsb.mpg.de/index.php/s/4nomcDH8EGwbzNi
mkdir <IPNet directory>/experiments
Put the downloaded weights in <IPNet directory>/experiments/

#-------------------------------------------------------------------------------
#Download animations from AIST++ or AMASS and extract each frame pose parameters to .npz files
Download dance animations from AIST++ here https://google.github.io/aistplusplus_dataset/download.html and extract poses for each frame to a file