AMASS preprocess

• In this repository, we provide the code for preprocessing and visulization of the Amass dataset.
• Preprocessing includes (1) saving parameters into .h5py format, (2) reduce the frame-per-second to 30hz.
• Also, we introduce how to rotate the motion for the augmentation. Refer to Rotate_hdf5.ipynb

Pre-requisites

1. Download ALL "SMPL-X Gender Specific" body information from the official webpage of AMASS dataset.
2. Unzip all *.tar.bz2 files.
3. Prepare Conda.

Installation

1. Prepare clean and new conda environment with Python3.7. Below the environment name is "amass".

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

2. Clone this Repository.

git clone https://github.com/cotton-ahn/amass_preprocess
cd amass_preprocess

3. Install Human_Body_Prior

git clone https://github.com/nghorbani/human_body_prior
cd human_body_prior
pip install -r requirements.txt

• (Tested on 23.12.2021) If you do until here, you would get the error of ERROR: No matching distribution found for torch==1.8.2.
• Check your CUDA environment and install PyTorch properly. Below example is for cuda 11.0. Refer HERE.

conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0 -c pytorch

• Modify requirements.txt. The pytorch version should be same as the one you installed
• And run pip install -r requirements.txt again in the folder of human_body_prior.
• If you get error regarding gcc and boost, try to install it. i.e, sudo apt-get install libboost-all-dev...
• And install the human_body_prior.

cd ${This Repository}/human_body_prior
python setup.py develop

4. Install PyRender

• Follow the description in Link

5. Download Gender-Specific SMPL-X models.

• Download SMPL-X with "removed head bun" for AMASS dataset.
• Put SMPL-X into folder of ${This Repository}/body_models/smplx, as below tree.

├── body_models/
│   ├── smplx/
│     ├── female/
│     ├── male/
│     ├── neutral/
│     ├── version.txt
└── README.md

6. Install body_visualizer

cd ${This Repository}
git clone https://github.com/nghorbani/body_visualizer
cd body_visualizer
pip install -r requirements.txt
python setup.py develop

7. Install other Dependencies

pip install h5py matplotlib 

8. Install jupyter notebook and add the activated conda environment.

pip install jupyter notebook
python -m ipykernel install --user --name=amass

How to use

1. Run Preprocess_hdf5.ipynb to save downloaded .npz data into .hdf5

• properly change amass_dir and h5py_dir.

2. Refer to Visualize_hdf5.ipynb to see how the saved hdf5 file work.
3. Refer to Rotate_hdf5.ipynb to see how to rotate the augment the given motion