A Scanner Deeply: Predicting Gaze Heatmaps On Visualizations Using Crowdsourced Eye Movement Data.

Area of Science:

• Human-Computer Interaction
• Computer Vision
• Cognitive Science

Background:

• Eye movement analysis is crucial for understanding visual perception in data visualization.
• Existing eye-tracking methods lack a standardized approach, hindering large-scale data collection.
• A comprehensive review of 30 prior works established a taxonomy for eye-movement data collection.

Purpose of the Study:

• To develop a scalable, low-cost eye-tracking method for data visualization research.
• To propose a novel virtual eye-tracker model (SCANNER DEEPLY) for generating gaze heatmaps from visualization images.
• To analyze eye movement patterns to understand user comprehension of visualized data structures.

Main Methods:

• A review of 30 eye-tracking studies categorized by tracker technology, image stimulus, and collection methodology.
• Implementation of a webcam-based eye-tracking system using data visualization images as stimuli.
• Development of a convolutional neural network model (SCANNER DEEPLY) trained on approximately 12,000 crowdsourced eye-movement samples.

Main Results:

• Successful development and validation of the SCANNER DEEPLY virtual eye-tracker model.
• Demonstrated the model's ability to generate gaze heatmaps for visualization images.
• Comparison of SCANNER DEEPLY with existing models (DVS, Salicon-trained network) showing competitive performance.

Conclusions:

• Webcam-based eye-tracking enables large-scale, crowdsourced data collection for visualization research.
• SCANNER DEEPLY provides an efficient computational model for predicting visual attention on data visualizations.
• The study contributes a novel dataset of visualization images and eye-movement data for future research.