Nymeria SMPL Preprocessor

A script to convert MVNX motion capture data to SMPL format.

demo.mp4

Overview

Converts MVNX files from the Nymeria dataset's data_xdata_mvnx directory to SMPL parameters and saves them to data_smpl_from_xdata_mvnx.

Directory Structure

nymeria_smpl_preprocessor/
├── preprocess_mvnx_to_smpl.py  # Main script
├── mvnx_to_smpl/               # Support library
│   └── core/
│       ├── articulate/         # SMPL body model utilities
│       └── paths.py            # Path configuration
└── README.md

Usage

# Default (with local pose computation, recommended)
python preprocess_mvnx_to_smpl.py

# Custom paths
python preprocess_mvnx_to_smpl.py \
    --input-dir /path/to/data_xdata_mvnx \
    --output-dir /path/to/data_smpl_from_xdata_mvnx

# Without local poses (not recommended)
python preprocess_mvnx_to_smpl.py --no-use-articulate

Output Format

The following files are generated in each sequence directory:

• smpl_data.npz: Numeric arrays only (numpy 1.x/2.x compatible)

  • global_poses: Global rotation matrices (N, 24, 3, 3)
  • local_poses: Local rotation matrices (N, 24, 3, 3)
  • root_positions: Root positions (N, 3)

• smpl_data.json: Metadata

  • joint_names: List of joint names
  • metadata: Frame rate and other info
  • num_frames: Number of frames
  • *_shape: Shape of each array

NumPy Compatibility

The npz files contain only numeric arrays, with metadata stored separately in JSON. This avoids pickle compatibility issues between numpy 1.x and 2.x.

Dependencies

• Python 3.8+
• PyTorch
• NumPy (1.x or 2.x)
• tqdm

Related Projects

• AITViewer: SMPL Visualization tool
• Nymeria Dataset: Motion capture data from Aria glasses + MVN suit