Visual Saliency

Visual saliency asks where people look and which regions are most important, supporting applications such as scene understanding, assistive vision, and safety. Methods range from biologically inspired contrast and motion models to deep networks that predict gaze maps, detect salient objects, or explain a model’s decisions; newer approaches condition attention on language, task goals, or scene semantics. Evaluation relies on eye-tracking and crowdsourced fixation benchmarks, object masks, and tests of whether highlighted regions faithfully explain model behavior. A central challenge is that attention varies with motion, goals, and individual or demographic differences, while many benchmarks and metrics still favor static scenes or central viewing bias.