GazeBase: Longitudinal Eye-Movement Recordings from 322 Subjects

GazeBase consists of 12,334 monocular eye-movement recordings captured from 322 college-aged subjects over 37 months. Subjects completed seven tasks per session, including fixation, saccade, reading, video viewing, and gaming tasks, using an EyeLink 1000 tracker at 1,000 Hz. The dataset was created by Henry Griffith for research in eye movement biometrics and machine learning applications.

Use Cases

• Developing biometric identification models based on longitudinal eye-movement patterns.
• Analyzing gaze behavior during reading tasks.
• Training classifiers for eye movement events (e.g., saccades, fixations) using instrument-provided labels.
• Studying gaze dynamics during free viewing of cinematic video.
• Exploring gaze-driven interaction patterns from the gaming task.

Strengths

• Large-scale longitudinal design with 322 subjects and 12,334 recordings.
• High-quality data collection using a 1,000 Hz EyeLink 1000 tracker with calibration protocols.
• Includes a battery of seven distinct behavioral tasks for varied context.
• Provides eye movement classification labels and pupil area data for a subset.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Data is monocular (left eye only), limiting binocular analysis.
• Participant pool is limited to college-aged subjects, which may introduce demographic bias.

Provenance

Source

Henry Griffith

Collection Method

Data collected using an EyeLink 1000 eye tracker with a calibration and validation protocol before each task.

Time Range

Recordings conducted over a 37-month period.