Datasets:

bkowshik
/

eeg-mi-datasets

Tags:

eeg

motor-imagery

brain-computer-interface

License:

Dataset card Files Files and versions

xet

Community

The Dataset Viewer has been disabled on this dataset.

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.

29 datasets · 815 subjects · 1,891 sessions · 633 GB in 4,051 files · 10 original hosts

At a glance
Datasets
Verification
Loading: BNCI2014_001 · BNCI2014_002 · BNCI2014_004 · BNCI2015_001 · Beetl2021_A · Brandl2020 · Cho2017 · Dreyer2023 · Farabbi2020 · GrosseWentrup2009 · GuttmannFlury2025_MI · Jia2019 · Kaya2018 · Kumar2024 · Lee2019_MI · Leeuwis2021 · Liu2024 · Pan2023 · Pan2025 · PhysionetMI · Rozado2015 · Shin2022 · Stieger2021 · Vagaja2023 · Wang2025 · Weibo2014 · Yang2025 · Zhou2016 · Zhou2020
Citation

At a glance

60% of the bytes are one dataset: Stieger2021 is 377 GB and holds 62 of the 815 subjects (8%).
25 GB holds 449 subjects (55%), across the 15 datasets that take under 0.1 GB per subject: BNCI2014_001, BNCI2014_002, BNCI2014_004, Dreyer2023, Jia2019, Kaya2018, Leeuwis2021, Liu2024, Pan2023, Pan2025, PhysionetMI, Rozado2015, Shin2022, Vagaja2023 and Zhou2016.
5 datasets do not allow modified copies to be shared: BNCI2014_001, BNCI2014_002 and BNCI2014_004 (CC BY-ND 4.0; 001 and 004 are BCI Competition IV data sets 2a and 2b), and BNCI2015_001 and Brandl2020 (CC BY-NC-ND 4.0). A filtered or epoched copy of them may not be distributed. The other 24 allow it, with attribution (CC BY 4.0, ODC-By 1.0) or without conditions (CC0 1.0).
7 datasets keep the names MOABB caches them under, mostly their host's numeric file IDs with no extension: Jia2019, Kumar2024, Leeuwis2021, Liu2024, Pan2023, Pan2025 and Shin2022, 303 files in all. Searching the Files tab for an upstream file name will not find them.
10 hosts served the originals: BNCI Horizon, Zenodo, figshare, Harvard Dataverse, DataverseNL, GigaDB, OSF, PhysioNet, DepositOnce and Hugging Face. Files were checked against MD5, SHA-1, SHA-256, CRC-32 or S3 ETag checksums; see Verification.
3 to 128 EEG channels, from BNCI2014_004 to GrosseWentrup2009, sampled at 160 Hz (PhysionetMI) to 1000 Hz, across the 24 datasets the verification loaded. Recordings come in at least eight formats: MAT, EDF, BDF, EEGLAB, BrainVision, XDF, NumPy and CSV.
9 of 106 motor imagery datasets in MOABB could not be fetched by a script on 7 Oct 2026: a 404, four Mendeley 403s, two licenses that must be accepted in code, a Harvard Dataverse 400 and a loader error. None of them is in this mirror.

Datasets

#	Dataset	Subjects	Sessions	Size	Source	License
1	BNCI2014_001	9	18	0.78 GB	BNCI Horizon 001-2014 (BCI Competition IV, data set 2a)	CC BY-ND 4.0, licensor: Institute for Knowledge Discovery, Graz University of Technology
2	BNCI2014_002	14	14	0.89 GB	BNCI Horizon 002-2014	CC BY-ND 4.0, licensor: Institute for Knowledge Discovery, Graz University of Technology
3	BNCI2014_004	9	45	0.48 GB	BNCI Horizon 004-2014 (BCI Competition IV, data set 2b)	CC BY-ND 4.0, licensor: Institute for Knowledge Discovery, Graz University of Technology
4	BNCI2015_001	12	28	1.83 GB	BNCI Horizon 001-2015	CC BY-NC-ND 4.0, licensor: Institute for Knowledge Discovery, Graz University of Technology
5	Beetl2021_A	3	3	0.45 GB	figshare `finalMI`, subjects 1–3 extracted from its single archive, and the final-phase labels	CC BY 4.0
6	Brandl2020	16	16	11.4 GB	DepositOnce, TU Berlin	CC BY-NC-ND 4.0
7	Cho2017	52	52	10.7 GB	GigaDB 100295	CC BY 4.0
8	Dreyer2023	87	87	5.5 GB	OSF, MOABB's BIDS copy of Zenodo	CC BY 4.0
9	Farabbi2020	12	36	2.5 GB	Zenodo	CC BY 4.0
10	GrosseWentrup2009	10	10	7.8 GB	Zenodo	CC BY 4.0
11	GuttmannFlury2025_MI	31	63	26.2 GB	MOABB's Zenodo copy of Synapse, motor imagery part	CC0 1.0
12	Jia2019	15	15	1.3 GB	figshare, the 30 experiment 1 files MOABB loads	CC BY 4.0
13	Kaya2018	7	17	0.64 GB	figshare, the 17 CLA recordings MOABB loads	CC0 1.0
14	Kumar2024	18	108	4.5 GB	Zenodo, its single archive	CC BY 4.0
15	Lee2019_MI	54	108	65 GB	GigaDB 100542, motor imagery files only	CC0 1.0
16	Leeuwis2021	55	55	5.1 GB	DataverseNL version 1.0, the EEG files	CC BY 4.0
17	Liu2024	50	50	0.46 GB	figshare, the EDF archive and two BIDS tables	CC BY 4.0
18	Pan2023	14	28	0.74 GB	Harvard Dataverse version 1.3	CC0 1.0
19	Pan2025	10	20	0.80 GB	Harvard Dataverse version 1.0	CC0 1.0
20	PhysionetMI	109	109	3.6 GB	PhysioNet eegmmidb 1.0.0, all 14 runs	ODC-By 1.0
21	Rozado2015	30	30	2.4 GB	Harvard Dataverse, extracted from its two RAR archives	CC0 1.0
22	Shin2022	10	10	0.35 GB	figshare, its single archive	CC BY 4.0
23	Stieger2021	62	598	377 GB	figshare	CC BY 4.0
24	Vagaja2023	26	26	1.9 GB	Zenodo, the motor imagery recordings extracted from `GROUPS.zip`	CC BY 4.0
25	Wang2025	15	30	3.8 GB	Hugging Face Jiaheng-Wang/ZJU-MI-EEG at `e60d6a7`, MI4 part	ODC-By 1.0
26	Weibo2014	10	10	4.3 GB	Harvard Dataverse, extracted from its three archives	CC0 1.0
27	Yang2025, two-class part	51	153	74.6 GB	figshare+, extracted from its single archive	CC BY 4.0
28	Zhou2016	4	12	0.13 GB	MOABB's Zenodo BIDS copy of figshare	CC BY 4.0
29	Zhou2020	20	140	17.6 GB	MOABB's Zenodo copy of IEEE DataPort	CC BY 4.0

Verification

Each file was checked as it was mirrored against the size and, where its host lists one, the checksum: an MD5, SHA-1 or SHA-256, or an archive member's CRC-32. After each commit, the Hub's SHA-256 was checked against the verified download.

On 8 Oct 2026, a fresh machine with no Hugging Face token restored every file of the 11 datasets added that day (Beetl2021_A, Farabbi2020, Jia2019, Kumar2024, Leeuwis2021, Liu2024, Pan2023, Pan2025, Shin2022, Vagaja2023 and Wang2025) and subjects 1–2 of 13 others. It checked each file again and loaded every one of those subjects with MOABB, with downloads blocked.

Loading

Every dataset loads the same way with MOABB at commit 555d14a, to be released as 1.8.0. The 18 datasets mirrored before 8 Oct 2026 also load with MOABB 1.7.

pip install "moabb @ git+https://github.com/NeuroTechX/moabb@555d14aebc1f1175e8476faeebabbfcafc522aba"

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

Each example below loads subject 1, except Vagaja2023's, whose subjects keep their upstream numbers. 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()

Beetl2021_A

Subjects 1–3 of the competition's final phase, extracted where MOABB extracts them from figshare's finalMI archive: under finalMI/S1/ for subject 1, five training runs with their labels (training/race1_padsData.npy, training/race1_padsLabel.npy, …) and ten testing runs (testing/race6_padsData.npy to testing/race15_padsData.npy). The testing runs' labels, final_MI_label.txt, come from figshare's labels article. The archive's S4 and S5 belong to Beetl2021_B and are left out. MOABB reads it from MNE_DATASETS_BEETL_PATH. For all 3 subjects, use allow_patterns="Beetl2021_A/*".

import os

from huggingface_hub import snapshot_download

patterns = ["Beetl2021_A/*/final_MI_label.txt", "Beetl2021_A/*/finalMI/S1/*"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns)
os.environ["MNE_DATASETS_BEETL_PATH"] = os.path.join(path, "Beetl2021_A")

import moabb
from moabb.datasets import Beetl2021_A

moabb.set_download_provider("upstream")
data = Beetl2021_A().get_data(subjects=[1])

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()

Farabbi2020

The per-subject zips MOABB downloads from Zenodo, 01.zip to 12.zip, each holding a subject's three sessions, and the record's chanlocs.locs. MOABB unpacks each zip beside it on first load, so download into a folder of your own. For all 12 subjects, use allow_patterns="Farabbi2020/*".

import os

from huggingface_hub import snapshot_download

patterns = ["Farabbi2020/*/chanlocs.locs", "Farabbi2020/*/01.zip"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns, local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_FARABBI2020_PATH"] = os.path.join(path, "Farabbi2020")

import moabb
from moabb.datasets import Farabbi2020

moabb.set_download_provider("upstream")
data = Farabbi2020().get_data(subjects=[1])

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()

Jia2019

The experiment 1 recordings, two MAT files per subject (exp1-S1-left.mat and exp1-S1-right.mat upstream), stored as MOABB stores them, under their figshare file IDs in files/. MOABB's loader maps each subject to its IDs: 14185838 and 14185841 for subject 1. For all 15 subjects, use allow_patterns="Jia2019/*".

import os

from huggingface_hub import snapshot_download

patterns = ["Jia2019/*/files/14185838", "Jia2019/*/files/14185841"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns)
os.environ["MNE_DATASETS_JIA2019_PATH"] = os.path.join(path, "Jia2019")

import moabb
from moabb.datasets import Jia2019

moabb.set_download_provider("upstream")
data = Jia2019().get_data(subjects=[1])

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()

Kumar2024

Zenodo's single 4.5 GB archive Online_Offline_Race.zip, under the name MOABB gives it, content. It holds every subject, and MOABB asks for it on every load, so it is needed even for one subject. MOABB unpacks it in the same folder on first load, so download into a folder of your own.

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Kumar2024/*", local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_KUMAR2024_PATH"] = os.path.join(path, "Kumar2024")

import moabb
from moabb.datasets import Kumar2024

moabb.set_download_provider("upstream")
data = Kumar2024().get_data(subjects=[1])

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()

Leeuwis2021

The EEG files, four CSV files per subject (calibration and three feedback runs), stored as MOABB stores them, under their DataverseNL file IDs in api/access/datafile/. MOABB numbers the subjects 1–55 in the order of their upstream numbers, which skip some, and its loader maps each subject to its IDs: 100063, 100065, 100033 and 100008 for subject 1. For all 55 subjects, use allow_patterns="Leeuwis2021/*".

import os

from huggingface_hub import snapshot_download

patterns = [f"Leeuwis2021/*/datafile/{id}" for id in (100063, 100065, 100033, 100008)]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns)
os.environ["MNE_DATASETS_LEEUWIS2021_PATH"] = os.path.join(path, "Leeuwis2021")

import moabb
from moabb.datasets import Leeuwis2021

moabb.set_download_provider("upstream")
data = Leeuwis2021().get_data(subjects=[1])

Liu2024

The three files MOABB downloads from figshare, under their figshare file IDs in files/: the archive of every subject's EDF recording, and the BIDS electrodes and events tables. MOABB unpacks the archive beside it on first load, so download into a folder of your own. The example downloads all 50 subjects, since one archive holds them.

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Liu2024/*", local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_LIU2024_PATH"] = os.path.join(path, "Liu2024")

import moabb
from moabb.datasets import Liu2024

moabb.set_download_provider("upstream")
data = Liu2024().get_data(subjects=[1])

Pan2023

One MAT file per session, S<n>D1.mat and S<n>D2.mat upstream, stored as MOABB stores them, under their Harvard Dataverse file IDs in api/access/datafile/. MOABB's loader maps each subject to its IDs: 7574475 and 7574479 for subject 1. For all 14 subjects, use allow_patterns="Pan2023/*".

import os

from huggingface_hub import snapshot_download

patterns = ["Pan2023/*/datafile/7574475", "Pan2023/*/datafile/7574479"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns)
os.environ["MNE_DATASETS_PAN2023_PATH"] = os.path.join(path, "Pan2023")

import moabb
from moabb.datasets import Pan2023

moabb.set_download_provider("upstream")
data = Pan2023().get_data(subjects=[1])

Pan2025

One MAT file per session, S<n>D1.mat and S<n>D2.mat upstream, stored as MOABB stores them, under their Harvard Dataverse file IDs in api/access/datafile/. MOABB's loader maps each subject to its IDs: 11704410 and 11704404 for subject 1. For all 10 subjects, use allow_patterns="Pan2025/*".

import os

from huggingface_hub import snapshot_download

patterns = ["Pan2025/*/datafile/11704410", "Pan2025/*/datafile/11704404"]
path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns=patterns)
os.environ["MNE_DATASETS_PAN2025_PATH"] = os.path.join(path, "Pan2025")

import moabb
from moabb.datasets import Pan2025

moabb.set_download_provider("upstream")
data = Pan2025().get_data(subjects=[1])

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()

Shin2022

figshare's single archive study-2022-data-anonymized.zip, under its figshare file ID in files/. It holds every subject; MOABB extracts the subject it loads beside it, so download into a folder of your own.

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Shin2022/*", local_dir="eeg-mi-datasets")
os.environ["MNE_DATASETS_SHIN2022_PATH"] = os.path.join(path, "Shin2022")

import moabb
from moabb.datasets import Shin2022

moabb.set_download_provider("upstream")
data = Shin2022().get_data(subjects=[1])

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()

Vagaja2023

The motor imagery recordings from Zenodo's GROUPS.zip, extracted where MOABB extracts them: Embodied/SUB03/SUB03_MI.vhdr with its .eeg and .vmrk for subject 3, and so on, under Embodied/ or Control/ by group. The archive's rest and embodiment recordings, which MOABB does not read, are left out. MOABB keeps the upstream subject numbers, 3, 5–10, 12 and 14–31, so the example loads subject 3. For all 26 subjects, use allow_patterns="Vagaja2023/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Vagaja2023/*/SUB03/*")
os.environ["MNE_DATASETS_VAGAJA2023_PATH"] = os.path.join(path, "Vagaja2023")

import moabb
from moabb.datasets import Vagaja2023

moabb.set_download_provider("upstream")
data = Vagaja2023().get_data(subjects=[3])

Wang2025

The MI4 part of the authors' Hugging Face repository at the revision MOABB's links served when mirrored, in the folder where MOABB keeps it: sub-01/s1_calibration.mat, s1_feedback.mat, s2_calibration.mat and s2_feedback.mat for subject 1, one calibration and one feedback file per session. For all 15 subjects, use allow_patterns="Wang2025/*".

import os

from huggingface_hub import snapshot_download

path = snapshot_download("bkowshik/eeg-mi-datasets", repo_type="dataset", allow_patterns="Wang2025/*/sub-01/*")
os.environ["MNE_DATASETS_WANG2025_PATH"] = os.path.join(path, "Wang2025")

import moabb
from moabb.datasets import Wang2025

moabb.set_download_provider("upstream")
data = Wang2025().get_data(subjects=[1])

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 works listed 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
Beetl2021_A: Wei, X., Faisal, A. A., Grosse-Wentrup, M., Gramfort, A., Chevallier, S., Jayaram, V., Jeunet, C., Bakas, S., Ludwig, S., Barmpas, K., Bahri, M., Panagakis, Y., Laskaris, N., Adamos, D. A., Zafeiriou, S., Duong, W. C., Gordon, S. M., Lawhern, V. J., Śliwowski, M., Rouanne, V., & Tempczyk, P. (2022). 2021 BEETL competition: advancing transfer learning for subject independence & heterogenous EEG data sets. arXiv, 2202.12950. https://doi.org/10.48550/arXiv.2202.12950
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
Farabbi2020: Farabbi, A., Ghiringhelli, F., Mainardi, L., Sanches, J. M., Moreno, P., Santos-Victor, J., Figueiredo, P., & Vourvopoulos, A. (2020). Motor-imagery EEG dataset during robot-arm control [Data set]. Zenodo. https://doi.org/10.5281/zenodo.5882500
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
Jia2019: Jia, T. (2019). EEG data of motor imagery for stroke [Data set]. figshare. https://doi.org/10.6084/m9.figshare.7636301
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
Kumar2024: Kumar, S., Alawieh, H., Racz, F. S., Fakhreddine, R., & Millán, J. del R. (2024). Transfer learning promotes acquisition of individual BCI skills. PNAS Nexus, 3(2), pgae076. https://doi.org/10.1093/pnasnexus/pgae076
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
Leeuwis2021: Leeuwis, N., Paas, A., & Alimardani, M. (2021). Psychological and cognitive factors in motor imagery brain computer interfaces (Version 1.0) [Data set]. DataverseNL. https://doi.org/10.34894/Z7ZVOD
Liu2024: Liu, H., Wei, P., Wang, H., Lv, X., Duan, W., Li, M., Zhao, Y., Wang, Q., Chen, X., Shi, G., Han, B., & Hao, J. (2024). An EEG motor imagery dataset for brain computer interface in acute stroke patients. Scientific Data, 11, 131. https://doi.org/10.1038/s41597-023-02787-8
Pan2023:
- Pan, L., Wang, K., Xu, L., Sun, X., Yi, W., Xu, M., & Ming, D. (2023). Riemannian geometric and ensemble learning for decoding cross-session motor imagery electroencephalography signals. Journal of Neural Engineering, 20(6), 066011. https://doi.org/10.1088/1741-2552/ad0a01
- Pan, L. (2023). A cross-session motor imagery EEG dataset (Version 1.3) [Data set]. Harvard Dataverse. https://doi.org/10.7910/DVN/251NOW
Pan2025: Pan, L. (2025). Cross-session motor imagery EEG dataset (Version 1.0) [Data set]. Harvard Dataverse. https://doi.org/10.7910/DVN/GH74ZG
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
Shin2022: Shin, H., Suma, D., & He, B. (2022). Closed-loop motor imagery EEG simulation for brain-computer interfaces. Frontiers in Human Neuroscience, 16, 951591. https://doi.org/10.3389/fnhum.2022.951591
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
Vagaja2023: Vagaja, K., & Vourvopoulos, A. (2023). Electrophysiological signals of embodiment and MI-BCI training in VR [Data set]. Zenodo. https://doi.org/10.5281/zenodo.8086086
Wang2025: Wang, J., Yao, L., & Wang, Y. (2025). Enhanced online continuous brain-control by deep learning-based EEG decoding. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 33, 2834–2846. https://doi.org/10.1109/TNSRE.2025.3591254
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

Downloads last month: 1,473

Total file size:

633 GB