[Preprint] FSDAM — Few-Shot Driver Attention Modeling

We are pleased to share our latest research project, FSDAM (Few-Shot Driver Attention Modeling), developed through a collaboration between Texas Tech University, Purdue University, and Towson University.

The project is led by Kaiser Hamid in collaboration with Can Cui, Khandakar Ashrafi Akbar, Ziran Wang, and Nade Liang.

FSDAM studies not only where drivers look, but also why their attention shifts. The framework jointly predicts spatial driver attention and generates structured reasoning explanations using only a small number of annotated examples. It decomposes attention into interpretable components including scene context, current focus, anticipated next focus, and causal reasoning.

The project combines a spatial prediction branch with a language-based reasoning branch, while using training-time vision–language alignment to inject semantic priors without increasing inference complexity.

Project page: FSDAM