Eye-Gaze Augmentation in Eye-Tracking Data
Georg Heller · 2024
Over the past decade, deep learning has achieved unprecedented successes across various application domains, driven by large-scale datasets. However, specific fields, such as healthcare, inherently face challenges like data scarcity and imbalance. Additionally, datasets may be largely inaccessible due to privacy concerns or lack of data-sharing incentives. These challenges highlight the importance of generative modeling and data augmentation in these domains.In this context, this study explores a machine learning-based approach for generating synthetic eye-tracking data. Eye tracking technology measures eye movements, positions, and points of gaze, providing insights into visual attention and behavior. We investigate a novel application of variational autoencoders (VAEs) for this purpose. Specifically, a VAE model is trained to generate image-based representations of eye-tracking outputs, known as scanpaths. Our results validate that the VAE model can generate plausible outputs from a limited dataset. Finally, it is empirically demonstrated that this approach can be used as a mechanism for data augmentation to improve performance in classification tasks.