Skip to content

Latest commit

 

History

History
 
 

smpl

README.MD

Installing the AMASS dependencies

To use the AMASS dataset, you need to download and set up the required data and models. This includes specifying the storage paths for the AMASS dataset and SMPL-H models and ensuring the dependencies are installed. Follow the instructions below to complete the setup.

Note

this currently works only with editable installations!

Additional dependencies

The retargeting pipeline builds on the SMPL models, which require PyTorch. It is recommended to install the CPU version of PyTorch to avoid conflicts with JAX.

To install the CPU version of PyTorch, run:

pip install torch --extra-index-url https://download.pytorch.org/whl/cpu

Once PyTorch is installed, you can install the rest of the additional dependencies, including SMPL-related packages. Note that this will install the GPU version of PyTorch if you haven't explicitly installed the CPU version above:

pip install -e '.[smpl]'

Download the AMASS Dataset and SMPL-H Model

1. Setting the Storage Directory

Before downloading the datasets and models, specify the directory where you’d like to store them:

mkdir <path to folder storing the amass datasets>
loco-mujoco-set-amass-path --path <path to folder storing the amass datasets>
mkdir <path to folder storing the smpl models>
loco-mujoco-set-smpl-model-path --path <path to folder storing the smpl models>

For example, to install the datasets in your home folder, you can run:

mkdir $HOME/amass
loco-mujoco-set-amass-path --path $HOME/amass
mkdir $HOME/smpl
loco-mujoco-set-smpl-model-path --path $HOME/smpl

Also, set the path to which the converted datasets (after retargeting) will be stored. If you already specified a general caching directory for LocoMuJoCo using loco-mujoco-set-all-caches, you can skip this step. But if you want to store the converted AMASS datasets in a different directory, you can set it using:

loco-mujoco-set-conv-amass-path --path <path to converted amass data>

For example, to set the path into the home directory, you can run:

loco-mujoco-set-conv-amass-path --path $HOME/amass_conv

2. Downloading the AMASS Dataset

  1. Visit the AMASS website.
  2. Register and download the datasets you wish to use.
  3. Select the SMPL-H G version of each dataset.
  4. Extract and place them in the directory specified above, e.g., $HOME/amass.

3. Downloading the SMPL-H Models

  1. Go to the MANO website.
  2. Register and download the following:
    • Extended SMPL+H model (includes the SMPL-H model w/o hands).
    • Models & Code (includes the hand models).
  3. Extract and place them in the directory specified above, e.g., $HOME/smpl.

Generating the SMPL-H Model

To combine the body and hand models into a single SMPLH_NEUTRAL.PKL file, follow one of the methods below:

Automatic Method (Recommended; Requires Anaconda)

To resolve dependency issues with deprecated packages, we provide an automated script. This script uses a Conda environment to manage dependencies and ensure proper installation.

  1. Ensure Anaconda is installed.
  2. Copy and paste the script somewhere.
  3. Run the script: 
    chmod u+x install_smplh.sh   # this is using the original name of the script
./install_smplh.sh

This script will:

  • Create a Python 3.10 environment.
  • Install all required dependencies.
  • Generate the SMPLH_NEUTRAL.PKL file and place it in /path/to/smpl/models.
  • Finally, delete the created Python environment and dependencies.

Manual Method (Without Conda)

If you don’t have Conda installed or prefer a manual setup, follow these steps. This method only supports Python 3.10.

  1. Create a Virtual Environment:

    python3.10 -m venv smplh_model_conversion_env
source smplh_model_conversion_env/bin/activate  # On Windows: smplh_model_conversion_env\Scripts\activate

  2. Install Required Dependencies:

    pip install 'numpy<1.23.0' chumpy tqdm pyyaml

  3. Run the Script:

    cd path/to/loco-mujoco
python loco_mujoco/smpl/generate_smplh_model.py --smpl-conf-file loco_mujoco/smpl/conf_paths.yaml

  4. Deactivate the Environment:

    deactivate

Test the datasets

You can test the SMPL model by running the script in examples/replay_datasets/:

python smpl_example.py

WARNING! The script assumes that you have downloaded the DanceDB dataset. You can try the script with different datasets by specifying a different dataset file