braindecode/NeuroPose-emg2pose

NeuroPose — emg2pose regression baseline

Figure 6 of Liu et al. (2021), CC BY.

Meta's released regression_neuropose.ckpt from the emg2pose benchmark, rehosted with braindecode's own parameter names so it loads directly:

from braindecode.models import NeuroPose

model = NeuroPose.from_pretrained("braindecode/NeuroPose-emg2pose")

16-channel sEMG at 2 kHz in, 20 joint angles per sample out — (batch, n_chans, n_times) to (batch, n_times, n_outputs).

Provenance

These are not retrained weights. They are the authors' released checkpoint with only the parameter names rewritten: upstream stores every block in a single nn.Sequential, while braindecode splits them across encoder / resnet / decoder. Loading the original .ckpt into braindecode.models.NeuroPose and running both implementations on the same input gives bit-identical outputs (max absolute difference 0.0, 6,354,903 parameters in both).

The Hub checkpoint uses Braindecode-native parameter names; upstream naming is handled once during publication rather than by the runtime model.

Architecture

Liu et al.'s encoder / ResNet / decoder, in the configuration emg2pose used for its published baseline: encoder widths 32/128/256 pooling (10,2) (8,2) (4,4) over (time x electrodes), five residual blocks of three conv groups, decoder 128/32/1 upsampling (10,4) (8,4) (4,2).

Training data

The data that shaped these weights. The ids are the datasets: field of this card, which braindecode reads into model.card_data.datasets when you call from_pretrained (MOABB class names for MOABB datasets, OpenNeuro / NEMAR accessions, Hub ids otherwise). nm/on ids are datasets on NEMAR (on = NEMAR's copy of the OpenNeuro dataset with the same number). Rows marked private: are data the authors did not release; they are not in datasets:.

datasets: id	Dataset	Stage	Modality
nm000281	emg2pose: sEMG (sEMG-RD wristbands, 2 kHz, 16 ch/hand) + motion-capture hand pose, 193 users, 370 h (Salter et al. 2024, NeurIPS D&B)	training	EMG

• experiment regression_neuropose (Liu et al. 2021 architecture re-trained by Meta on emg2pose; not Liu et al.'s own data). Supervised on the emg2pose train split (Table 3: 158 users, 23 stages, 250.9 h); val split monitored for checkpointing/early stopping (monitor val_loss); held-out user / stage / user-stage test sets not trained on.
• Sources: arXiv:2412.02725 (HTML) Table 3; https://github.com/facebookresearch/emg2pose README 'Downloading Pre-trained Checkpoints'; https://github.com/facebookresearch/emg2pose config/base.yaml.

License and attribution

Weights are Meta's, released under CC BY-NC-SA 4.0 — non-commercial, share-alike. That license carries over here unchanged. The UmeTrack hand model used for the forward kinematics behind the labels is CC BY-NC 4.0.

• Salter et al. (2024), emg2pose: A Large and Diverse Benchmark for Surface Electromyographic Hand Pose Estimation, NeurIPS Datasets and Benchmarks. arXiv:2412.02725
• Liu, Zhang, Gowda (2021), NeuroPose: 3D Hand Pose Tracking using EMG Wearables, The Web Conference 2021. doi:10.1145/3442381.3449890

Cite Braindecode

If you use this braindecode port in scientific work, cite the software on Zenodo, doi:10.5281/zenodo.17699192, and the original model and datasets cited above.

@software{braindecode,
  author = {Aristimunha, Bruno and
            Guetschel, Pierre and
            Wimpff, Martin and
            Gemein, Lukas and
            Rommel, Cedric and
            Banville, Hubert and
            Sliwowski, Maciej and
            Wilson, Daniel and
            Brandt, Simon and
            Gnassounou, Théo and
            Paillard, Joseph and
            {Junqueira Lopes}, Bruna and
            Sedlar, Sara and
            Moreau, Thomas and
            Chevallier, Sylvain and
            Gramfort, Alexandre and
            Schirrmeister, Robin Tibor},
  title = {Braindecode: toolbox for decoding raw electrophysiological brain data
           with deep learning models},
  url = {https://github.com/braindecode/braindecode},
  doi = {10.5281/zenodo.17699192},
  publisher = {Zenodo},
  license = {BSD-3-Clause},
}