brainomni-base-pretrained

Weights of BrainOmni base (lm_dim 512, 16 heads, 12 blocks) with its frozen tokenizer, for braindecode.models.BrainOmni, converted from the authors' release. The classification head is not pretrained (seeded random init); fine-tune or linear-probe before use.

from braindecode.models import BrainOmni
model = BrainOmni.from_pretrained("braindecode/brainomni-base-pretrained", chs_info=raw.info["chs"], n_outputs=2)

chs_info must carry sensor positions (EEG) and coil orientations (MEG); the chs_info in config.json (19 EEG channels, 10-20) is only a default. Input is expected at 256 Hz, preprocessed as in the authors' code.

Source and conversion

• Source: OpenTSLab/BrainOmni at revision 9a4d3c70495370397ccfbfd6d2496f25647545a5, file base/BrainOmni.pt (sha256 435db24e57a55df05aa7e16355def7b7ecbedb22aa1ec16063e7d14efd2386d0), MIT licence.
• convert_brainomni_checkpoints.py (in this repository) renames the keys to braindecode's, drops the pretraining-only mask predictor, stores the RoPE cache as (cos, sin) pairs with zero sine (the released cache holds cosines only and the released code uses it as loaded) and writes config.json, model.safetensors and pytorch_model.bin with save_pretrained.
• The converted model's outputs equal braindecode's loading of the original file (max-abs difference 0.0, float32 and bfloat16).
• Requires a braindecode version newer than 1.8.1.

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
nm000229	MEG-MASC (Gwilliams et al. 2023, Sci Data 10:862); 27 English speakers, KIT 208-ch (paper ref [24])	pretraining, pretraining	MEG
Armeni2022	MEG-Narrative-Dataset in the BrainOmni paper (Armeni et al. 2022, Sci Data 9:278, 10-hour within-participant narrative MEG); CTF 275-ch, 3 subjects (paper ref [2])	pretraining, pretraining	MEG
OMEGA	OMEGA: The Open MEG Archive (Niso et al. 2016, NeuroImage 124); 444 healthy + 200 patients, resting state (paper ref [45])	pretraining, pretraining	MEG
CamCAN	Cam-CAN CC700 (Taylor et al. 2017, NeuroImage 144); called 'CC700' in the paper; ~700 subjects, tasks + rest, Elekta 306-ch (paper ref [75])	pretraining, pretraining	MEG
on002680	Go-Nogo: go-nogo categorization and detection task (Delorme) (paper ref [16])	pretraining, pretraining	EEG
on002721	MusicEEG: an EEG dataset recorded during affective music listening (Daly et al.) (paper ref [13])	pretraining, pretraining	EEG
on003555	HFO: EEG recordings of pediatric patients with epilepsy based on the 10-20 system (Cserpan et al.) (paper ref [11])	pretraining, pretraining	EEG
on003682	AversiveMEG: model-based aversive learning in humans (Wise et al.) (paper ref [82])	pretraining, pretraining	MEG
on003775	SRM resting-state EEG (Hatlestad-Hall et al.) (paper ref [27])	pretraining, pretraining	EEG
on004107	MIND DATA (Weisend et al.); median-nerve stimulation (paper ref [81])	pretraining, pretraining	MEG
on004148	RestCog: a test-retest resting and cognitive state EEG dataset (Wang et al.) (paper ref [80])	pretraining, pretraining	EEG
on004078	SMN4Lang: a synchronized multimodal neuroimaging dataset to study brain language processing (Wang et al.) (paper ref [79])	pretraining, pretraining	MEG
ds004186	HBN EO/EC task (Alexander et al.; Healthy Brain Network resting eyes-open/eyes-closed) (paper ref [41])	pretraining, pretraining	EEG
ds004212	THINGS-MEG (Hebart et al.) (paper ref [30])	pretraining, pretraining	MEG
on004276	ASWR-MEG: auditory single word recognition in MEG (Gaston et al.) (paper ref [21])	pretraining, pretraining	MEG
ds004330	ImageLine: spatiotemporal neural dynamics of object recognition for natural images and line drawings (MEG) (Singer et al.) (paper ref [68])	pretraining, pretraining	MEG
on004357	Features-EEG (Grootswagers et al.) (paper ref [23])	pretraining, pretraining	EEG
on004796	PEARL-Neuro: Polish EEG, Alzheimer's risk-genes, lifestyle and neuroimaging database (Dzianok & Kublik) (paper ref [51])	pretraining, pretraining	EEG
on005241	NeuroMorph: high-temporal resolution MEG dataset for morpheme-based linguistic analysis (Rodriguez et al.) (paper ref [57])	pretraining, pretraining	MEG
Kymata-SOTO	Kymata SOTO language dataset: an electro-magnetoencephalographic dataset for natural speech processing (Yang et al. 2025); simultaneous MEG 306-ch + EEG 64-ch (paper ref [87]); modality MEG+EEG	pretraining, pretraining	MEG
on005505	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 1 (Shirazi et al.) (paper ref [60])	pretraining, pretraining	EEG
on005506	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 2 (Shirazi et al.) (paper ref [61])	pretraining, pretraining	EEG
on005507	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 3 (Shirazi et al.) (paper ref [62])	pretraining, pretraining	EEG
on005508	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 4 (Shirazi et al.) (paper ref [63])	pretraining, pretraining	EEG
on005509	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 5 (Shirazi et al.) (paper ref [64])	pretraining, pretraining	EEG
on005510	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 6 (Shirazi et al.) (paper ref [65])	pretraining, pretraining	EEG
on005511	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 7 (Shirazi et al.) (paper ref [66])	pretraining, pretraining	EEG
on005512	HBN-EEG: Healthy Brain Network (HBN) EEG - Release 8 (Shirazi et al.) (paper ref [67])	pretraining, pretraining	EEG
on005620	Awakening: a repeated awakening study exploring complexity measures to capture dreaming during propofol sedation (Bajwa et al.) (paper ref [6])	pretraining, pretraining	EEG

• Stage 1: BrainTokenizer (SEANet+RVQ) trained on the Appendix H corpus (85% train partition)
• Stage 2: BrainOmni masked-token pretraining on the same Appendix H corpus, tokenizer frozen (shipped inside this checkpoint)
• Sources: arXiv:2505.18185v3 (HTML), Sec. 4 'Pretraining Datasets'; arXiv:2505.18185v3 bibliography (accessions).

Citation

@inproceedings{xiao2025brainomni,
  title     = {BrainOmni: A Brain Foundation Model for Unified EEG and MEG Signals},
  author    = {Xiao, Q. and Cui, Z. and Zhang, C. and Chen, S. and Wu, W. and
               Thwaites, A. and Woolgar, A. and Zhou, B. and Zhang, C.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2025},
  note      = {arXiv:2505.18185},
}

@article{aristimunha2025braindecode,
  title   = {Braindecode: a deep learning library for raw electrophysiological data},
  author  = {Aristimunha, Bruno and others},
  journal = {Zenodo},
  year    = {2025},
  doi     = {10.5281/zenodo.17699192},
}

License

MIT, as the original BrainOmni release.

Current availability of a cited source

ds004186 is retained because the BrainOmni paper lists it as pretraining data. On 2026-11-09 it was not anonymously readable on OpenNeuro and had no public NEMAR mirror; this card does not imply current public access.