Fixation Importance Prediction

Description: Gaze data is used for a wide variety of tasks as it conveys various information. A typical used type of eye-movement is the fixation. Usually every fixation is used for a given task, sometimes weighted by duration. However, not every fixation is necessarily important for the task. The goal of this thesis is to predict the importance of a fixation from the gaze sequence. For this, you will use an existing gaze dataset and semi-automatically label the fixation importance (this depends on the used dataset and the task the gaze data is used for). You then need to design, implement and train a (neural network) model for importance prediction. Afterwards, you should analyse if this model can be used to improve the performance of a task which uses gaze data by filtering fixations according to the predicted importance.

Supervisor: Florian Strohm

Distribution: 40% Implementation, 20% data preparation, 40% Analysis and Evaluation

Requirements: Good Python skills; interest in deep learning and eye-tracking