Selective Visual Attention Decoding Dataset KU Leuven

If using this dataset, please cite the following paper and the current Zenodo repository.

[1] Y. Yao, W. D. Swaef, S. Geirnaert and A. Bertrand, "EEG-Based Decoding of Selective Visual Attention in Superimposed Videos," in IEEE Journal of Biomedical and Health Informatics, doi: 10.1109/JBHI.2025.3580261.

The associated code is available at: https://github.com/YYao-42/EEG-based_SVAD

Overview

This dataset contains electroencephalogram (EEG) and eye-tracking data collected during a selective visual attention experiment. The research was conducted at the Department of Electrical Engineering (ESAT), KU Leuven.

Dataset Summary:

• Participants: 19 young adults with normal or corrected-to-normal vision
• Stimuli: single-object natural videos and superimposed-object videos
• Total Duration: 19 subjects x 28 min (single-object) + 19 subjects x 38 min (superimposed-object)
• Data Types: 64-channel EEG, 4-channel electrooculogram (EOG) and gaze coordinates

Experimental Design

(Further details can be found in [1].)

Video Stimuli

Please refer to the uploaded exp_video_creation.pdf for a detailed illustration of the video creation procedure.

• Fourteen one-shot videos are used to create seven video pairs. The object in each video is centered.
• For each pair, the longer video is truncated to match the length of the shorter one. The videos are presented as follows: single video for 0-120 seconds, two-second fade-in transition at 120-122 seconds, then fully superimposed videos from 122 seconds to the end.
• Each edited video begins with instruction frames, directing participants to focus on whichever object is shown during the initial single-object phase (0-120 seconds).
• Each video pair produced two trials, with participants instructed to attend to a different object in each trial.
• As a result, the dataset includes:
  • Single-object data: 19 subjects × 28 minutes
  • Superimposed-object data: 19 subjects × 38 minutes

Data Acquisition

EEG Recording:

• Device: BioSemi ActiveTwo
• Sampling rate: 2048 Hz
• Channels: 64 EEG + 4 EOG (Electrode position coordinates for the BioSemi 64-channel cap are available here.)

Eye Tracking:

• Device: Pupil Labs NEON eye tracker
• Sampling rate: 200 Hz (downsampled to 30 Hz to match the video frame rate)
• Output: Each sample includes four columns: horizontal gaze coordinate, vertical gaze coordinate, saccade indicator (1 = saccade, 0 = no saccade), and blink indicator (1 = blink, 0 = no blink). Coordinates are provided in pixels, matching the 1920×1080 video resolution.

Contents

YouTube Videos

• Videos can be downloaded using the provided download_video.py script, which requires the yt-dlp tool. (If any videos have been removed by the uploader, please contact us.)
• All videos are resampled to 30 Hz and resized to 1920×1080 resolution.
• Video URLs and timestamps are included in the script (see lines 37–50). Users may choose to download the videos in alternative ways using this information.

EEG and Eye Gaze Data

• Data (in the zip file) is organized into 19 folders, one per subject. Raw EEG data is provided in both EEGLAB .set (metadata) and .fdt (raw data) formats, compatible with tools like MNE. Gaze data is provided in .npy format.
• Single-Object Data (0–120 seconds of each video)
  • EEG files are named using the corresponding video ID (e.g., 01.set for Video 01).
  • Gaze data follow the same naming, with _gaze appended (e.g., 01_gaze.npy).
• Superimposed-Object Data (122 seconds to video end; 2-second fade-in excluded)
  • EEG files are named as video1_video2_attended, indicating the superimposed videos and the attended video. For example, 01_13_01.set corresponds to a trial where Video 01 and Video 13 were superimposed, and participants attended to Video 01.
  • Gaze data follow the same convention with the _gaze postfix (e.g., 01_13_01_gaze.npy)

Acknowledgement

This research is funded by the Research Foundation - Flanders (FWO) project No G081722N, junior postdoctoral fellowship fundamental research of the FWO (for S. Geirnaert, No. 1242524N), the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement No 101138304), Internal Funds KU Leuven (project IDN/23/006), and the Flemish Government (AI Research Program).

We also thank the participants for their time and effort in the experiments.

Contact Information

Executive researcher: Yuanyuan Yao, yuanyuan.yao@kuleuven.be

Led by: Prof. Alexander Bertrand, alexander.bertrand@kuleuven.be