Predicting Visual Fixations.

Area of Science:

• Cognitive Science
• Neuroscience
• Computer Vision

Background:

• Human eye movements are continuous decisions reflecting unconscious and conscious processes.
• Eye movement trajectories offer insights into visual and cognitive functions.
• Predicting where humans look is a key challenge in understanding visual attention.

Purpose of the Study:

• To review advances in predicting human eye movements.
• To establish consistent methods for evaluating and comparing predictive models.
• To assess the contribution of different mechanisms in gaze prediction.

Main Methods:

• Utilizing probabilistic models for a unified approach to fixation prediction.
• Translating diverse saliency and scanpath models into a common framework.
• Employing information gain as a metric for model comparison.

Main Results:

• Probabilistic models enable consistent comparison across static and video saliency, and scanpath prediction.
• The framework allows quantification of contributions from different predictive factors.
• Selection of informative examples for model comparison is facilitated.

Conclusions:

• A universal scale of information gain is a powerful tool for inspecting predictive mechanisms.
• This approach aids in designing experiments to understand gaze determination.
• Unified model evaluation advances the study of continuous decision-making in visual attention.