I’m thrilled to announce that I have successfully defended my PhD thesis!
Thesis Title
“A Gaze into the Art World: Predicting Visual Attention Using Deep Learning”
Abstract
Cultural heritage, a cornerstone of societal and historical identity, necessitates a comprehensive understanding and preservation to effectively bridge past and present. Paintings, spanning from prehistoric cave art to modern contemporary pieces, form a significant part of this heritage. They are subject to continuous scrutiny on historical, artistic, and technological fronts. As art embodies the zenith of human intelligence and creativity, the study of human cognitive behavior in response to paintings has gained considerable attraction.
This thesis delves into the exploration of human visual behavior through the lens of eye-tracking technologies. Given the scarcity of publicly accessible data, we have amassed a rich dataset that encapsulates various art schools and movements, featuring paintings with diverse degrees of realism and abstraction.
We have put forth several deep neural network-based methodologies for gaze trajectory and saliency prediction, aiming to model visual attention behavior. Our focus has been on enhancing the model’s representational capacity within the painting domain, given its unique properties compared to natural scenes. To this end, we have employed techniques such as unsupervised domain adaptation and self-supervised learning. Our models incorporate several inductive biases based on visual attention theory.
The objective of this thesis is twofold: to contribute to research by offering a public art and eye-tracking dataset, and to introduce novel deep learning-based modeling methods tailored for the domain of paintings.
Thesis Preview
Links
Acknowledgments
This journey would not have been possible without the support of my supervisors, collaborators at PRISME Laboratory, and my family. I’m also grateful to all the colleagues and institutions I collaborated with during these four years.
On to new adventures!