eeg-mi-datasets

EEG motor imagery datasets

Unmodified copies of public EEG motor imagery datasets in the MOABB cache layout. Files extracted from an upstream archive keep their bytes; Weibo2014's also take the names MOABB gives them. Zhou2016's saved Zenodo record is the one generated file; see its section.

• Datasets
• Loading: BNCI2014_001 · BNCI2014_002 · BNCI2014_004 · BNCI2015_001 · Brandl2020 · Cho2017 · Dreyer2023 · GrosseWentrup2009 · GuttmannFlury2025_MI · Kaya2018 · Lee2019_MI · PhysionetMI · Rozado2015 · Stieger2021 · Weibo2014 · Yang2025 · Zhou2016 · Zhou2020
• Citation

Datasets

Loading

Every dataset loads the same way with MOABB 1.7:

1. Download the files you need with snapshot_download, narrowed by allow_patterns.
2. Point MOABB at them through its MNE_DATASETS_<SIGN>_PATH environment variable.
3. Call moabb.set_download_provider("upstream") so MOABB uses the local files instead of contacting NEMAR.

Each example below loads subject 1. The pattern for all of a dataset's subjects is given in its text.

BNCI2014_001

Each subject's two sessions as MOABB downloads them from BNCI Horizon: A01T.mat (training) and A01E.mat (evaluation) for subject 1, and so on, plus the dataset's description.pdf. MOABB reads every BNCI dataset from MNE_DATASETS_BNCI_PATH. For all 9 subjects, use allow_patterns="BNCI2014_001/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="BNCI2014_001/*/A01*")
os.environ["MNE_DATASETS_BNCI_PATH"] = os.path.join(path, "BNCI2014_001")

import moabb
from moabb.datasets import BNCI2014_001

moabb.set_download_provider("upstream")
data = BNCI2014_001(subjects=[1]).get_data()

BNCI2014_002

Each subject's training and evaluation files as MOABB downloads them from BNCI Horizon: S01T.mat and S01E.mat for subject 1, and so on, plus the dataset's description.pdf. MOABB loads both files as one session. MOABB reads every BNCI dataset from MNE_DATASETS_BNCI_PATH, so point it at one BNCI dataset's folder at a time. For all 14 subjects, use allow_patterns="BNCI2014_002/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="BNCI2014_002/*/S01*")
os.environ["MNE_DATASETS_BNCI_PATH"] = os.path.join(path, "BNCI2014_002")

import moabb
from moabb.datasets import BNCI2014_002

moabb.set_download_provider("upstream")
data = BNCI2014_002(subjects=[1]).get_data()

BNCI2014_004

Each subject's files as MOABB downloads them from BNCI Horizon: B01T.mat holds sessions 1–3 and B01E.mat sessions 4–5 for subject 1, and so on, plus the dataset's description.pdf. MOABB reads every BNCI dataset from MNE_DATASETS_BNCI_PATH. For all 9 subjects, use allow_patterns="BNCI2014_004/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="BNCI2014_004/*/B01*")
os.environ["MNE_DATASETS_BNCI_PATH"] = os.path.join(path, "BNCI2014_004")

import moabb
from moabb.datasets import BNCI2014_004

moabb.set_download_provider("upstream")
data = BNCI2014_004(subjects=[1]).get_data()

BNCI2015_001

One file per session as MOABB downloads them from BNCI Horizon: S01A.mat and S01B.mat for subject 1, and so on; subjects 8–11 also have S08C.mat to S11C.mat. The dataset's description.pdf comes with them. MOABB reads every BNCI dataset from MNE_DATASETS_BNCI_PATH. For all 12 subjects, use allow_patterns="BNCI2015_001/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="BNCI2015_001/*/S01*")
os.environ["MNE_DATASETS_BNCI_PATH"] = os.path.join(path, "BNCI2015_001")

import moabb
from moabb.datasets import BNCI2015_001

moabb.set_download_provider("upstream")
data = BNCI2015_001(subjects=[1]).get_data()

Brandl2020

One MAT file per subject, pp1.mat to pp16.mat, and the montage mnt.mat that MOABB loads with every subject, as MOABB downloads them from DepositOnce. The README and the two band and interval files beside them upstream are left out. For all 16 subjects, use allow_patterns="Brandl2020/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=["Brandl2020/*/mnt.mat", "Brandl2020/*/pp1.mat"])
os.environ["MNE_DATASETS_BRANDL2020_PATH"] = os.path.join(path, "Brandl2020")

import moabb
from moabb.datasets import Brandl2020

moabb.set_download_provider("upstream")
data = Brandl2020(subjects=[1]).get_data()

Cho2017

GigaDB's mat_data/s<NN>.mat, one file per subject, along with the readmes, questionnaire results and trial sequence. MOABB reads it from MNE_DATASETS_GIGADB_PATH. For all 52 subjects, use allow_patterns="Cho2017/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Cho2017/*/s01.mat")
os.environ["MNE_DATASETS_GIGADB_PATH"] = os.path.join(path, "Cho2017")

import moabb
from moabb.datasets import Cho2017

moabb.set_download_provider("upstream")
data = Cho2017(subjects=[1]).get_data()

Dreyer2023

The files MOABB downloads from OSF: one zip per subject, plus the manifest and release files MOABB checks for. MOABB unpacks each zip beside it on first load, so download into a folder of your own. For all 87 subjects, use allow_patterns="Dreyer2023/*".

import os

from huggingface_hub import snapshot_download

# Release files (all names but sub-*.zip) and subject 1.
patterns = ["Dreyer2023/*/[!s]*", "Dreyer2023/*/stimuli.zip", "Dreyer2023/*/sub-01.zip"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns, local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_DREYER2023_PATH"] = os.path.join(path, "Dreyer2023")

import moabb
from moabb.datasets import Dreyer2023

moabb.set_download_provider("upstream")
data = Dreyer2023(subjects=[1]).get_data()

GrosseWentrup2009

Each subject's EEGLAB recording as MOABB downloads it from Zenodo: subject1.set and its data file subject1.fdt, and so on. MOABB reads it from MNE_DATASETS_MUNICHMI_PATH. For all 10 subjects, use allow_patterns="GrosseWentrup2009/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="GrosseWentrup2009/*/subject1.*")
os.environ["MNE_DATASETS_MUNICHMI_PATH"] = os.path.join(path, "GrosseWentrup2009")

import moabb
from moabb.datasets import GrosseWentrup2009

moabb.set_download_provider("upstream")
data = GrosseWentrup2009(subjects=[1]).get_data()

GuttmannFlury2025_MI

The per-subject zips of the motor imagery part that MOABB downloads from Zenodo, each holding one to three of a subject's sessions. MOABB unpacks each zip beside it on first load, so download into a folder of your own. For all 31 subjects, use allow_patterns="GuttmannFlury2025_MI/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="GuttmannFlury2025_MI/*/S01.zip", local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_GUTTMANNFLURY2025-MI_PATH"] = os.path.join(path, "GuttmannFlury2025_MI")

import moabb
from moabb.datasets import GuttmannFlury2025_MI

moabb.set_download_provider("upstream")
data = GuttmannFlury2025_MI(subjects=[1]).get_data()

Kaya2018

The 17 classical left hand, right hand and passive recordings (CLA) that MOABB loads, one per session, under their figshare names: CLA-SubjectA-160108-3St-LRHand.mat and so on. MOABB numbers subjects A–F and J as 1–7. The collection's other paradigms are not included. For all 7 subjects, use allow_patterns="Kaya2018/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Kaya2018/*/CLA-SubjectA-*")
os.environ["MNE_DATASETS_KAYA2018_PATH"] = os.path.join(path, "Kaya2018")

import moabb
from moabb.datasets import Kaya2018

moabb.set_download_provider("upstream")
data = Kaya2018(subjects=[1]).get_data()

Lee2019_MI

The motor imagery files MOABB downloads from GigaDB, one MAT file per session, along with GigaDB's readmes, questionnaire results and MD5 list. The ERP, SSVEP and artifact recordings in the same GigaDB dataset are not included. For all 54 subjects, use allow_patterns="Lee2019_MI/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Lee2019_MI/*/sess*_subj01_*")
os.environ["MNE_DATASETS_LEE2019-MI_PATH"] = os.path.join(path, "Lee2019_MI")

import moabb
from moabb.datasets import Lee2019_MI

moabb.set_download_provider("upstream")
data = Lee2019_MI(subjects=[1]).get_data()

PhysionetMI

All 14 EDF runs of each subject, S001/S001R01.edf to S001/S001R14.edf for subject 1, in the folder where MOABB and MNE's eegbci keep them. These cover both MOABB's default imagery runs and the executed runs (PhysionetMI(executed=True)). The .edf.event files are left out, since the EDF annotations hold the same events. MOABB reads it from MNE_DATASETS_EEGBCI_PATH. For all 109 subjects, use allow_patterns="PhysionetMI/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="PhysionetMI/*/S001/*")
os.environ["MNE_DATASETS_EEGBCI_PATH"] = os.path.join(path, "PhysionetMI")

import moabb
from moabb.datasets import PhysionetMI

moabb.set_download_provider("upstream")
data = PhysionetMI(subjects=[1]).get_data()

Rozado2015

The 60 recordings from Dataverse's two RAR archives, extracted where MOABB extracts them: extracted/1/exp1/experiment.xdf and extracted/1/exp2/experiment.xdf for subject 1, and so on. MOABB finds these and needs no RAR tool. For all 30 subjects, use allow_patterns="Rozado2015/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Rozado2015/*/extracted/1/*")
os.environ["MNE_DATASETS_ROZADO2015_PATH"] = os.path.join(path, "Rozado2015")

import moabb
from moabb.datasets import Rozado2015

moabb.set_download_provider("upstream")
data = Rozado2015(subjects=[1]).get_data()

Stieger2021

One MAT file per session, as MOABB downloads them from figshare. For all 62 subjects, use allow_patterns="Stieger2021/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Stieger2021/*/S1_*")
os.environ["MNE_DATASETS_STIEGER2021_PATH"] = os.path.join(path, "Stieger2021")

import moabb
from moabb.datasets import Stieger2021

moabb.set_download_provider("upstream")
data = Stieger2021(subjects=[1], sessions=[1]).get_data()

Weibo2014

The ten recordings from Dataverse's three archives, each extracted and renamed as MOABB does, from the participant's initials to subject_<n>.mat; the bytes are unchanged. The archives' shared ReadMe.txt is left out. For all 10 subjects, use allow_patterns="Weibo2014/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Weibo2014/*/subject_1.mat")
os.environ["MNE_DATASETS_WEIBO_PATH"] = os.path.join(path, "Weibo2014")

import moabb
from moabb.datasets import Weibo2014

moabb.set_download_provider("upstream")
data = Weibo2014(subjects=[1]).get_data()

Yang2025

The two-class part of figshare's single 65.6 GB archive, extracted in the folder where MOABB unpacks it:

• each session's raw recording (data.bdf) and events (evt.bdf) under sourcedata/2C dataset/sub-<NNN>/;
• the authors' preprocessed .mat files under derivatives/2C dataset_processeddata/;
• the release files.

MOABB skips the archive when it finds these files. The three-class part (11 people) and six JPEG images beside the two-class recordings are left out. For all 51 subjects, use allow_patterns="Yang2025/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Yang2025/*/sourcedata/2C dataset/sub-001/*")
os.environ["MNE_DATASETS_YANG2025_PATH"] = os.path.join(path, "Yang2025")

import moabb
from moabb.datasets import Yang2025

moabb.set_download_provider("upstream")
data = Yang2025(subjects=[1]).get_data()

Zhou2016

The per-subject zips of MOABB's BIDS copy on Zenodo, each holding a subject's three sessions, plus the BIDS release files and 16534752.json, the Zenodo record MOABB saves and reads offline. That record is the one file not copied byte for byte: it is saved without Zenodo's daily view and download counts, which MOABB never reads. MOABB unpacks each zip beside it on first load, so download into a folder of your own. For all 4 subjects, use allow_patterns="Zhou2016/*".

import os

from huggingface_hub import snapshot_download

# Release files (all names but sub-*.zip) and subject 1.
patterns = ["Zhou2016/*/[!s]*", "Zhou2016/*/sub-1.zip"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns, local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_ZHOU2016_PATH"] = os.path.join(path, "Zhou2016")

import moabb
from moabb.datasets import Zhou2016

moabb.set_download_provider("upstream")
data = Zhou2016(subjects=[1]).get_data()

Zhou2020

The per-subject zips MOABB downloads from Zenodo, each holding all seven of a subject's sessions. MOABB unpacks each zip beside it on first load, so download into a folder of your own. For all 20 subjects, use allow_patterns="Zhou2020/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Zhou2020/*/S01.zip", local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_ZHOU2020_PATH"] = os.path.join(path, "Zhou2020")

import moabb
from moabb.datasets import Zhou2020

moabb.set_download_provider("upstream")
data = Zhou2020(subjects=[1]).get_data()

Citation

Cite the paper for each dataset you use.

• BNCI2014_001:
• Tangermann, M., Müller, K.-R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K. J., Müller-Putz, G. R., Nolte, G., Pfurtscheller, G., Preissl, H., Schalk, G., Schlögl, A., Vidaurre, C., Waldert, S., & Blankertz, B. (2012). Review of the BCI Competition IV. Frontiers in Neuroscience, 6, 55. https://doi.org/10.3389/fnins.2012.00055
• Brunner, C., Leeb, R., Müller-Putz, G. R., Schlögl, A., & Pfurtscheller, G. (2008). BCI Competition 2008 – Graz data set A. Graz University of Technology. https://lampx.tugraz.at/~bci/database/001-2014/description.pdf
• BNCI2014_002: Steyrl, D., Scherer, R., Faller, J., & Müller-Putz, G. R. (2016). Random forests in non-invasive sensorimotor rhythm brain-computer interfaces: a practical and convenient non-linear classifier. Biomedical Engineering / Biomedizinische Technik, 61(1), 77–86. https://doi.org/10.1515/bmt-2014-0117
• BNCI2014_004:
• Leeb, R., Lee, F., Keinrath, C., Scherer, R., Bischof, H., & Pfurtscheller, G. (2007). Brain–computer communication: motivation, aim, and impact of exploring a virtual apartment. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 15(4), 473–482. https://doi.org/10.1109/TNSRE.2007.906956
• Tangermann, M., Müller, K.-R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K. J., Müller-Putz, G. R., Nolte, G., Pfurtscheller, G., Preissl, H., Schalk, G., Schlögl, A., Vidaurre, C., Waldert, S., & Blankertz, B. (2012). Review of the BCI Competition IV. Frontiers in Neuroscience, 6, 55. https://doi.org/10.3389/fnins.2012.00055
• BNCI2015_001: Faller, J., Vidaurre, C., Solis-Escalante, T., Neuper, C., & Scherer, R. (2012). Autocalibration and recurrent adaptation: towards a plug and play online ERD-BCI. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 20(3), 313–319. https://doi.org/10.1109/TNSRE.2012.2189584
• Brandl2020: Brandl, S., & Blankertz, B. (2020). Motor imagery under distraction — an open access BCI dataset. Frontiers in Neuroscience, 14, 566147. https://doi.org/10.3389/fnins.2020.566147
• Cho2017: Cho, H., Ahn, M., Ahn, S., Kwon, M., & Jun, S. C. (2017). EEG datasets for motor imagery brain-computer interface. GigaScience, 6(7), gix034. https://doi.org/10.1093/gigascience/gix034
• Dreyer2023: Dreyer, P., Roc, A., Pillette, L., Rimbert, S., & Lotte, F. (2023). A large EEG database with users' profile information for motor imagery brain-computer interface research. Scientific Data, 10, 580. https://doi.org/10.1038/s41597-023-02445-z
• GrosseWentrup2009: Grosse-Wentrup, M., Liefhold, C., Gramann, K., & Buss, M. (2009). Beamforming in noninvasive brain–computer interfaces. IEEE Transactions on Biomedical Engineering, 56(4), 1209–1219. https://doi.org/10.1109/TBME.2008.2009768
• GuttmannFlury2025_MI: Guttmann-Flury, E., Sheng, X., & Zhu, X. (2025). Dataset combining EEG, eye-tracking, and high-speed video for ocular activity analysis across BCI paradigms. Scientific Data, 12, 587. https://doi.org/10.1038/s41597-025-04861-9
• Kaya2018: Kaya, M., Binli, M. K., Ozbay, E., Yanar, H., & Mishchenko, Y. (2018). A large electroencephalographic motor imagery dataset for electroencephalographic brain computer interfaces. Scientific Data, 5, 180211. https://doi.org/10.1038/sdata.2018.211
• Lee2019_MI: Lee, M.-H., Kwon, O.-Y., Kim, Y.-J., Kim, H.-K., Lee, Y.-E., Williamson, J., Fazli, S., & Lee, S.-W. (2019). EEG dataset and OpenBMI toolbox for three BCI paradigms: an investigation into BCI illiteracy. GigaScience, 8(5), giz002. https://doi.org/10.1093/gigascience/giz002
• PhysionetMI:
• Schalk, G., McFarland, D. J., Hinterberger, T., Birbaumer, N., & Wolpaw, J. R. (2004). BCI2000: a general-purpose brain-computer interface (BCI) system. IEEE Transactions on Biomedical Engineering, 51(6), 1034–1043. https://doi.org/10.1109/TBME.2004.827072
• Schalk, G. (2009). EEG Motor Movement/Imagery Dataset (version 1.0.0). PhysioNet. https://doi.org/10.13026/C28G6P
• Goldberger, A., Amaral, L., Glass, L., Hausdorff, J., Ivanov, P. C., Mark, R., Mietus, J. E., Moody, G. B., Peng, C.-K., & Stanley, H. E. (2000). PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation, 101(23), e215–e220. https://doi.org/10.1161/01.CIR.101.23.e215
• Rozado2015: Rozado, D., Duenser, A., & Howell, B. (2015). Improving the performance of an EEG-based motor imagery brain computer interface using task evoked changes in pupil diameter. PLoS ONE, 10(3), e0121262. https://doi.org/10.1371/journal.pone.0121262
• Stieger2021: Stieger, J. R., Engel, S. A., & He, B. (2021). Continuous sensorimotor rhythm based brain computer interface learning in a large population. Scientific Data, 8, 98. https://doi.org/10.1038/s41597-021-00883-1
• Weibo2014: Yi, W., Qiu, S., Wang, K., Qi, H., Zhang, L., Zhou, P., He, F., & Ming, D. (2014). Evaluation of EEG oscillatory patterns and cognitive process during simple and compound limb motor imagery. PLoS ONE, 9(12), e114853. https://doi.org/10.1371/journal.pone.0114853
• Yang2025: Yang, B., Rong, F., Xie, Y., Li, D., Zhang, J., Li, F., Shi, G., & Gao, X. (2025). A multi-day and high-quality EEG dataset for motor imagery brain-computer interface. Scientific Data, 12, 488. https://doi.org/10.1038/s41597-025-04826-y
• Zhou2016: Zhou, B., Wu, X., Lv, Z., Zhang, L., & Guo, X. (2016). A fully automated trial selection method for optimization of motor imagery based brain-computer interface. PLoS ONE, 11(9), e0162657. https://doi.org/10.1371/journal.pone.0162657
• Zhou2020: Zhou, Q., Lin, J., Yao, L., Wang, Y., Han, Y., & Xu, K. (2021). Relative power correlates with the decoding performance of motor imagery both across time and subjects. Frontiers in Human Neuroscience, 15, 701091. https://doi.org/10.3389/fnhum.2021.701091