|
Description
|
Based on the paper: 'Gaze3P: Gaze-Based Prediction of User-Perceived Privacy'. Eye Tracking data collected with the EyeLink eye-tracker.
Type of data:
The Gaze3P dataset contains multimodal experimental data, combining three principal categories: - Eye-tracking data: Gaze coordinates, fixations, saccades and pupil size. Measures visual attention and behaviour - Subjective ratings: Privacy assessments on a scale from 1 to 7. Provides the ground-truth labels - Stimulus and experimental metadata: Images, attributes, task conditions and participant information. Provides context for interpreting gaze behaviour
The dataset contains gaze recordings from 100 participants viewing 1,000 natural images.
1. Eye-tracking data
This is the main behavioural data type.
It includes: - Gaze coordinates (X, Y), indicating where participants looked. - Timestamps, indicating when they looked at a particular location. - Fixations, representing periods when the eyes remain relatively stable. - Saccades, representing rapid movements between fixation points. - Pupil size, which can provide additional information about visual and cognitive responses. The original recordings are collected using an EyeLink eye tracker at a sampling rate of 2,000 Hz.
2. Subjective privacy ratings
Participants evaluate how private or safe they consider the information presented in an image. The ratings use a seven-point scale: - 1 = Very private. - 7 = Very safe. These ratings are particularly important because they serve as the reference labels for the machine-learning models.
3. Stimulus data
The visual stimuli originate from the VISPR (Visual Privacy) dataset. They contain natural images associated with different privacy-related attributes, including personal information, documents, health, employment, relationships and location. The images allow the researchers to investigate which visual elements attract attention and how this relates to perceived privacy.
------------------------------------------------------------------------------------
Context in which the data was collected:
The objective was to investigate whether privacy perception can be predicted implicitly through eye movements rather than relying exclusively on questionnaires.
The data was collected in a controlled laboratory environment. The main technical configuration reported in the paper includes: Parameter Configuration Eye tracker EyeLink Sampling frequency 2,000 Hz Screen resolution 1920 × 1080 pixels Screen dimensions 545 × 303 mm Eye-to-screen distance 700 mm Head stabilization Chin rest Calibration 13-point calibration
The researchers used two experimental tasks:
Task 1: Free-viewing
Participants viewed individual images and assessed their perceived privacy. The purpose was to observe natural visual attention without requiring participants to search for a specific object.
Task 2: Search task
Participants searched for a particular category of information within a visual arrangement of images. This task allowed researchers to investigate gaze behaviour during goal-directed visual attention.
------------------------------------------------------------------------------------
Structure of the dataset:
The original eye-tracking recordings are extracted using EyeLink Data Viewer and stored in .edf files.
The raw files contain: Variable Description Data type Timestamp Time at which a gaze measurement was recorded Numeric Gaze X Horizontal screen coordinate Numeric Gaze Y Vertical screen coordinate Numeric Fixation duration Duration of a fixation Numeric Saccade duration Duration of a saccade Numeric Saccade amplitude Distance between successive fixation points Numeric Saccade velocity Speed of eye movement Numeric Pupil size Measured pupil diameter Numeric
------------------------------------------------------------------------------------
Dataset usage:
The dataset can be used directly within the EyeLink software.
|
|
Related Publication
| Is Supplement To: Elfares, M., Reisert, P., Küsters, R., & Bulling, A. (2026). Gaze3P: Gaze-based prediction of user-perceived privacy. Proceedings on Privacy Enhancing Technologies, 2026(1), 6-25.doi: 10.56553/popets-2026-0002 |