EEG for natural-image recognition (VEP)

Description

This is the “EEG dataset for natural image recognition through visual stimuli” — a visual-evoked-potential (VEP) corpus collected by Nandan Tiwari, Shamama Anwar, and Vandana Bhattacharjee at the Department of Computer Science and Engineering, Birla Institute of Technology (Mesra/Ranchi, India) and published as a data descriptor in Data in Brief (Tiwari et al., 2025). The recordings capture scalp EEG while healthy adult volunteers viewed photographs of four natural-object categories — apple, car, flower, and human face — with the explicit aim of supporting EEG-based image classification, visual decoding, and image-reconstruction research (i.e. mapping brain responses back to the perceived object). All participants were pre-screened on the Vividness of Visual Imagery Questionnaire (VVIQ) so that the cohort had a documented, moderate-to-high capacity for mental imagery. The work was funded by SERB/ANRF (Anusandhan National Research Foundation, India) under the SUPRA scheme (file SPR/2022/000154).

Citation

• Paper DOI: 10.1016/j.dib.2025.111639
• Dataset DOI: 10.17632/g9shp2gxhy.2
• Reference: Tiwari, N., Anwar, S., & Bhattacharjee, V. (2025). EEG dataset for natural image recognition through visual stimuli. Data in Brief 60, 111639.

Catalog citation block

Tiwari, N., Anwar, S., & Bhattacharjee, V. (2025). EEG dataset for natural image recognition through visual stimuli. Data in Brief, 60, 111639. DOI: 10.1016/j.dib.2025.111639.

Dataset: Anwar, S., Tiwari, N., & Bhattacharjee, V. (2025). EEG Dataset for natural image recognition through Visual Stimuli (Version 2) [Data set]. Mendeley Data. DOI: 10.17632/g9shp2gxhy.2.

Download and access

Field	Value
Official source	https://data.mendeley.com/datasets/g9shp2gxhy/2
Access class	open
Size	TODO(confirm)
Format	csv, edf
BIDS	false
Mirrors	none recorded

Source & access

The data are hosted on Mendeley Data: “EEG Dataset for natural image recognition through Visual Stimuli”, version 2, DOI 10.17632/g9shp2gxhy.2 (https://data.mendeley.com/datasets/g9shp2gxhy/2). The accompanying peer-reviewed data descriptor is Tiwari, Anwar & Bhattacharjee (2025), Data in Brief vol. 60, article 111639, DOI 10.1016/j.dib.2025.111639 (open access, PMC12149563). The license is contested between the two sources: the Mendeley landing page states CC BY 4.0, while the Data in Brief article text states CC BY-NC. Treat the more restrictive CC BY-NC (non-commercial) as the safe assumption pending clarification. Ethics approval was issued by Birla Institute of Technology.

Data structure

The release is a plain folder hierarchy (not BIDS): a top-level VEP-DATA/ containing VVIQuestionnaire.pdf, Participant_info.csv, and one folder per image class (A/, C/, F/, P/), each split into two exemplar subfolders (e.g. A1/, A2/). Raw EEG is provided in both CSV and EDF formats. Each CSV carries a Timestamp column, EEG.Counter, EEG.Interpolated, the per-channel raw EEG samples (EEG.<sensor>, in microvolts), and Emotiv-derived per-channel band-power feature columns: Theta (4—8 Hz), Alpha (8—12 Hz), BetaL (12—16 Hz), BetaH (16—25 Hz), and Gamma (25—45 Hz). No preprocessed derivatives are provided — the data are raw.

License

• Name: CC-BY-4.0
• Note (verbatim from the directory): contested: CC BY 4.0 on the Mendeley landing page vs CC BY-NC in the Data in Brief article — the more restrictive (non-commercial) reading governs until the author resolves it (§13 item 15)
• Snippets on this site: no — the site ships no snippet or precomputed product from this dataset; lessons link the official access page.

Acquisition

Field	Value
Device	14-channel wireless saline-electrode headset (Emotiv EPOC X)
Device class	consumer
Channels	14
Sampling rate (Hz)	128
Online filters	raw and unfiltered per the authors; device front-end inferred from the hardware specification (≈ 0.16–43 Hz, 50/60 Hz notches)
Reference	TODO(confirm)
Mains frequency (Hz)	TODO(confirm)
Paradigms	passive-viewing
Participants	32
Population	32 adults 20–46 (24 M / 8 F), screened for visual-imagery vividness (VVIQ 40–72)
Clinical groups	none
Sessions	TODO(confirm)
Durations	~60-s sessions per image class; baseline image ~7 s

Acquisition

Recorded on the Emotiv EPOC X 14-channel wireless EEG headset using the EmotivPRO software for acquisition and annotation (Tiwari et al., 2025). The 14 saline-felt electrodes follow the Emotiv 10—20 montage: AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4 — frontally weighted, no Cz/Pz midline coverage, but with bilateral occipital electrodes (O1, O2) suited to the posterior visual evoked response. The device outputs at 128 Hz. The EPOC X streams at an effective 128 Hz (2048 Hz internal sampling, decimated) through a built-in 5th-order Sinc filter giving a hardware bandwidth of ~0.16—43 Hz, with hardware digital notch filters at both 50 Hz and 60 Hz; the descriptor itself does not restate the online filter/reference settings, so these are inferred from the EPOC X hardware specification rather than documented in the paper. Stimuli were presented on a 22-inch LCD monitor; the baseline (target) image was held for ~7 s and each image-class session ran ~60 s.

Participants

35 adults were enrolled; 32 completed the study and constitute the released dataset (Tiwari et al., 2025). The completing cohort was 24 male / 8 female, aged 20—46 years (mean ~34). All participants are healthy volunteers — there is no clinical group. Inclusion required a VVIQ score in the band 40 ≤ VVIQ ≤ 72 (16 items rated on a five-point scale), selecting for moderate-to-vivid visual imagery and excluding aphantasic and hyper-vivid extremes. A Participant_info.csv and the VVIQuestionnaire.pdf are bundled with the release.

Tasks / conditions

The paradigm is a passive natural-image viewing VEP task with four object classes labeled A = apple, C = car, F = flower, P = human face (Tiwari et al., 2025). Each class has two distinct exemplar images (A1/A2, C1/C2, F1/F2, P1/P2). Within a recording, the baseline/target image is embedded among randomly interleaved images: in Set 1 the target appears around 46—52 s after two random images, and in Set 2 around 52—58 s after three random images.

Used on this site

L7.5

Loader

Fetches only the stated subset. Recorded loader: TODO(confirm).

TODO(confirm) — the catalog records no loader for this dataset.

Caveats worth knowing

• Nominal 128 Hz output but ≈ 43 Hz hardware bandwidth and built-in 50/60 Hz notches — the Nyquist frequency is not the usable bandwidth.

Notable caveats

The authors state explicitly that this is raw EEG with no pre-processing or filtering — expect EOG/EMG artifacts and power-line noise from the recording environment (Tiwari et al., 2025). The dual 50/60 Hz hardware notches leave no usable content near the line frequencies. Note that the Emotiv “Gamma (25—45 Hz)” band-power column is computed up against this roll-off and should not be trusted as a clean gamma estimate.

Related datasets

• MultiPENG (EEG stream)
• ArEEG_Words
• EEGEmotions-27
• PhySF, Physiological Sense of Flow
• ERP CORE (Compendium of Open Resources and Experiments)

# Loader: TODO(confirm) — the catalog records no loader for this dataset.
# Download a small subset from the official source: https://data.mendeley.com/datasets/g9shp2gxhy/2