Dario Zanca (DIISM, University of Siena)
Mar 20, 2019 – 11:00 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
When eye-tracking devices are not a viable option, models of visual attention can be used to predict locations fixated by humans. We argue that the idea of a saliency map is ill-posed, and present models of full scanpaths instead. Three proposal in the framework of machanics will be discussed. First, bottom-up attention is modeled as local laws of motion which are derived from a few functional principles, including the brightness invariance. Second, it is shown how to include information coming from the inner activation of a convolutional neural network, in a first attempt of modeling a top-down signal. Third and finally, a general model is discussed which unifies previous proposals and extends them towards a more biologically founded theory of human attention.