Discover datasets

THINGS-EEG2 is a large EEG dataset from 10 participants who viewed 16,540 training images and 200 repeatedly presented test images from the THINGS database. Images were shown in rapid serial visual presentation at 5 Hz while 63-channel EEG was recorded at 1000 Hz across four sessions per subject. The dataset supports modeling of human visual object recognition and includes resting-state recordings, the authors' preprocessed epochs, and stimulus images.

Participants

10

Channels

63 (10-10)

Citations

88

Size

241 GB

Version

v1.1.0

Updated

Jul 5, 2026

EEG recordings from fifteen healthy participants who executed and imagined six sustained upper limb movements (elbow flexion/extension, forearm supination/pronation, hand open/close) plus a rest condition, in two separate sessions. The dataset supports decoding of single upper limb movements from the time-domain of low-frequency EEG. It is relevant to non-invasive control of motor neuroprostheses and robotic arms. This BIDS version was generated from the original BNCI Horizon 2020 data using MOABB.

Participants

15

Channels

61 (10-05)

Citations

93

HED

v8.4.0

Size

34.0 GB

Version

v1.0.3

Updated

Aug 18, 2026

The High-gamma dataset contains 128-channel EEG recordings from 14 healthy participants performing cue-based executed movements of the left hand, right hand and feet, plus rest. It was introduced to evaluate end-to-end decoding with deep convolutional neural networks against filter bank common spatial patterns (FBCSP), and to visualize the spectral power modulations these networks learn. This version is a BIDS-formatted derivative generated with MOABB.

Participants

14

Channels

128 (10-05)

Citations

78

HED

v8.4.0

Size

31.3 GB

Version

v1.0.3

Updated

Aug 18, 2026

Electroencephalographic recordings from 14 healthy participants performing a cue-guided, two-class motor imagery task (right hand versus feet) in a Graz-BCI paradigm with continuous visual feedback. Each participant completed a single session of 8 runs and 160 trials, recorded with 15 EEG channels at 512 Hz. The dataset was used to compare random forest and regularized LDA classifiers for sensorimotor rhythm brain-computer interfaces.

Participants

14

Channels

15 (other)

HED

v8.4.0

Size

1.37 GB

Version

v1.0.5

Updated

Oct 8, 2026

EEG recordings from 10 able-bodied participants, plus a feasibility recording from one participant with spinal cord injury, performing attempted lower-arm movements to control a continuous 2D cursor trajectory. Each of the able-bodied participants completed three sessions comprising offline calibration and online feedback with 50% and 100% EEG control. The dataset supports regression-based trajectory decoding, learning effects across sessions, and source-localization analyses for brain-computer interface research.

Participants

10

Channels

60 (10-05)

Citations

20

HED

v8.4.0

Size

7.38 GB

Version

v1.0.3

Updated

Aug 18, 2026

This dataset is a subset (Study N) of the bigP3BCI collection, comprising 16-channel EEG recordings from 8 participants with amyotrophic lateral sclerosis (ALS) using a P300-based speller with a 9x8 character grid. Each participant completed two sessions comparing dry and wet electrode setups. Target and non-target visual stimulus events are annotated with HED tags, and the data were recorded at 256 Hz with g.USBamp amplifiers.

Participants

8

Channels

16 (10-10)

Citations

2

HED

v8.4.0

Size

1.22 GB

Version

v1.0.4

Updated

Oct 8, 2026

This dataset contains EEG recordings from 8 healthy participants in Study E of the bigP3BCI collection, which used a 6x6 checkerboard character-grid P300 speller paradigm. Data were recorded from 16 EEG channels at 256 Hz with a g.USBamp amplifier. The recordings contain target and non-target visual stimulus events with HED annotations, and this version is a BIDS-formatted derivative of the PhysioNet bigP3BCI release, prepared through MOABB.

Participants

8

Channels

16 (10-10)

Citations

2

HED

v8.4.0

Size

518 MB

Version

v1.0.3

Updated

Oct 8, 2026

This dataset is Study K of the bigP3BCI collection, a P300-based brain-computer interface speller dataset. It contains 16-channel EEG recorded from 5 healthy participants across 2 sessions, using a 9x8 character grid with adaptive and checkerboard stimulus paradigms. Target and non-target events are annotated with HED tags. The data are a BIDS-formatted derivative of the PhysioNet bigP3BCI release, prepared with MOABB.

Participants

5

Channels

16 (10-10)

Citations

2

HED

v8.4.0

Size

790 MB

Version

v1.0.4

Updated

Oct 8, 2026

EEG recordings from eight people with amyotrophic lateral sclerosis (ALS) performing a visual P300 speller task with a 6x6 matrix. Participants copy-spelled seven predefined five-character words using row-column intensification while 8-channel EEG was recorded at 256 Hz. The dataset supports study of how attention and working memory relate to P300-based brain-computer interface performance in a clinical population.

Participants

8

Channels

8 (10-10)

Citations

89

HED

v8.4.0

Size

241 MB

Version

v1.0.3

Updated

Oct 8, 2026

EEG dataset of error-related potentials (ErrPs) recorded from six participants while they monitored a cursor moving autonomously toward a target, with 20% or 40% error probability. Participants had no control over the cursor, showing that error correlates can be observed and decoded when monitoring an external agent. Recordings were made with 64 channels at 512 Hz across 20 sessions per subject, and the data support single-trial error detection for brain-computer interfaces.

Participants

6

Channels

64 (10-10)

Citations

81

HED

v8.4.0

Size

5.21 GB

Version

v1.0.2

Updated

Aug 18, 2026