Visual Attention

Visual attention research asks where viewers look, how gaze changes with goals and scene content, and how selective vision can guide recognition and action. Methods include eye-tracking-based saliency maps and scanpath prediction, biologically inspired foveation, active perception with reinforcement learning, and attention mechanisms in vision transformers and vision-language models. Work spans visual-search, driving, medical-imaging, and multimodal benchmarks, with growing interest in using selective attention to reduce computation or improve grounding. A central challenge is that model attention maps do not always reflect human gaze or faithfully explain model decisions, while metrics can reward center bias or overlook task-specific behavior.