stac-mjx 🐀

stac-mjx is an implementation of the Stac algorithm for inverse kinematics on markerless motion capture data. It's written in MJX for hardware acceleration .

This is part of the Virtual Neurosceince Lab (VNL) project.

Installation

stac-mjx relies on many prerequisites, therefore we suggest installing in a new conda environment, using the provided environment.yaml: [Local installation before package is officially published]

1. Clone the repository git clone https://github.com/talmolab/stac-mjx.git and cd into it
2. Create and activate the stac-mjx-env environment:

conda env create -f environment.yaml
conda activate stac-mjx-env

Our rendering functions support multiple backends: egl, glfw, and osmesa. We show osmesa setup as it supports headless rendering, which is common in remote/cluster setups. To set up (currently on supported on Linux), execute the following commands sequentially:

sudo apt-get install libglfw3 libglew2.0 libgl1-mesa-glx libosmesa6 
conda install -c conda-forge glew 
conda install -c conda-forge mesalib 
conda install -c anaconda mesa-libgl-cos6-x86_64 
conda install -c menpo glfw3

Finally, set the following environment variables, and reactivate the conda environment:

conda env config vars set MUJOCO_GL=osmesa PYOPENGL_PLATFORM=osmesa
conda deactivate && conda activate base

To ensure all of the above changes are encapsulated in your Jupyter kernel, create a new kernel with:

conda install ipykernel
python -m ipykernel install --user --name stac-mjx-env --display-name "Python (stac-mjx-env)"

Usage

1. Update the .yaml files in config/ with the proper information (details WIP).

2. Run stac-mjx with its basic api: load_configs for loading configs and run_stac for the keypoint registration. Below is an example script, found in demos/use_api.ipynb. A CLI script is also provided at run_stac.py. Refer to hydra documention for formatting args to override configs.

   import stac_mjx 
   from pathlib import Path
   
   # Enable XLA flags if on GPU
   stac_mjx.enable_xla_flags()
   
   # Choose parent directory as base path for data files
   base_path = Path.cwd().parent
   
   # Load configs
   cfg = stac_mjx.load_configs(base_path / "configs")
   
   # Load data
   kp_data, sorted_kp_names = stac_mjx.load_data(cfg, base_path)
   
   # Run stac
   fit_path, ik_only_path = stac_mjx.run_stac(
    cfg,
    kp_data, 
    sorted_kp_names, 
    base_path=base_path
   )

3. Render the resulting data using mujoco_viz() (example notebook found in demos/viz_usage.ipynb):

   import stac_mjx
   
   import mediapy as media
   from pathlib import Path
   import os
   
   base_path = Path.cwd()
   cfg = stac_mjx.load_configs(base_path / "configs")
   
   stac_cfg, model_cfg = main.load_configs(stac_config_path, model_config_path)
   
   data_path = base_path / "demo_fit.p"
   n_frames = 250
   save_path = base_path / "videos/direct_render.mp4"
   
   # Call mujoco_viz
   frames = viz_stac(data_path, cfg, n_frames, save_path, start_frame=0, camera="close_profile", base_path=Path.cwd().parent)
   
   # Show the video in the notebook (it is also saved to the save_path)
   media.show_video(frames, fps=cfg.model.RENDER_FPS)

4. If the rendering is poor, it's likely that some hyperparameter tuning is necessary. (details WIP)