Hugging Face's logo

Datasets:
amathislab
/
ultra-mocap-emg-preprocessed

Modalities:
Tabular
Time-series
Formats:
csv
Size:
100K - 1M
Tags:
biomechanics
electromyography
timeseries
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

Agreement required before accessing this dataset

This repository contains EMG data derived from the CC BY 4.0-licensed ULTRA-MoCap dataset and prepared for use with MM-EvalKit.

By requesting access to this dataset, you acknowledge and agree that:

  1. These files are derived from the ULTRA-MoCap dataset and remain subject to the Creative Commons Attribution 4.0 International license.
  2. You will provide appropriate credit to the original ULTRA-MoCap authors, link to the license, and indicate if you share or modify these files.
  3. You understand that this is an independent preprocessing release for MM-EvalKit, not the official processed ULTRA-MoCap release.
  4. You are responsible for using the data in accordance with applicable laws, institutional policies, and ethical research practices.
  5. You will cite ULTRA-MoCap and MM-EvalKit or MuscleMimic in resulting publications.

Log in or Sign Up to review the conditions and access this dataset content.

ULTRA-MoCap EMG Preprocessed for MM-EvalKit

This repository contains 208 preprocessed surface EMG recordings from the 13 subjects in ULTRA-MoCap. The files use the directory and CSV format expected by MM-EvalKit.

Source

The files are derived from the official ULTRA-MoCap dataset, which is licensed under CC BY 4.0.

This is an independent derived release. It is not the official processed ULTRA-MoCap dataset and does not imply endorsement by the original authors.

Processing

The raw 2,000 Hz EMG recordings were processed per channel using:

  1. mean centering;
  2. fourth-order zero-phase Butterworth band-pass filtering at 20–500 Hz;
  3. full-wave rectification;
  4. Hampel outlier filtering;
  5. a centered RMS envelope;
  6. subject-level task-peak normalization; and
  7. interpolation onto the matching 100 Hz Processed IK time grid without extrapolation.

Task-peak normalization is used because ULTRA-MoCap does not provide separate maximum voluntary contraction recordings.

File format

Each trial is stored as:

PXX/<movement>/EMG_preprocessed/<trial>_preprocessed.csv

Each CSV contains:

  • time: seconds on the 100 Hz Processed IK timeline;
  • IM EMG4: biceps brachii;
  • IM EMG5: triceps brachii; and
  • IM EMG6: middle deltoid.

Use with MM-EvalKit

Download the repository:

hf auth login
hf download amathislab/ultra-mocap-emg-preprocessed \
  --repo-type dataset \
  --local-dir ultra-mocap-emg

Configure MM-EvalKit:

mm-evalkit paths set ultra-mocap /absolute/path/to/ultra-mocap-emg

The ULTRA-MoCap evaluation and rendering commands will discover the trial files under this directory.

Citation

Please cite the original dataset:

@article{fritsche2026ultra,
  title={ULTRA-MoCap: A multimodal IMU and sEMG dataset for upper body joint
    kinematics analysis},
  author={Fritsche, Oliver and Camacho, Steven and Hossain, Md Sanzid Bin and
    Halfpenny, Tyler and Arciniegas, Carlos and Dranetz, Joseph and Hadley,
    Dexter and Guo, Zhishan and Choi, Hwan},
  journal={Scientific Data},
  volume={13},
  number={1},
  pages={622},
  year={2026},
  publisher={Nature Publishing Group UK London}
}

When using these files with MM-EvalKit, also cite MuscleMimic as described in the MM-EvalKit README.

License

These derived files are distributed under CC BY 4.0, following the license of the source dataset. You must provide appropriate attribution, link to the license, and indicate whether changes were made.
Downloads last month
6
Total file size:
64.4 MB

Collection including amathislab/ultra-mocap-emg-preprocessed