Modeling eye movements with static heatmaps that represents topological probablity of attracting attention has a long history in the literature. Heatmaps are blurred fixations maps that only takes position into account. On the hand, Duration of fixation is something interseting that has not recieved enough attention. Something that has not been studied in depth in eye movement analysis is whether there is any agreement based on duration of fixation and relative position in stimuli. Is it possible to model duration of fixations? What would happen if we re-weight and normalize fixation maps before blurring using duration? Would it help model improve the performance in predicting static salaiency?
