Excited to share that our paper “Salypath: A Deep-Based Architecture For Visual Attention Prediction” has been published on arXiv!

The Paper

Salypath introduces a novel deep learning architecture for predicting visual attention - where humans look when viewing images. This work represents a significant step in my PhD research on understanding and modeling human gaze behavior.

Key Contributions

  • A unified deep architecture that predicts both saliency maps (where people look) and scanpaths (the sequence of eye movements)
  • Novel loss functions tailored for scanpath prediction
  • Evaluation on standard benchmarks showing competitive performance

Abstract

Understanding visual attention is crucial for many applications. In this work, we propose a deep learning-based approach that jointly addresses saliency prediction and scanpath generation, providing a more complete picture of human visual behavior.

This work was presented at ICIP 2021. I’m grateful to my collaborators and the PRISME Laboratory for their support.