Evaluation of Visual Saliency Models in Immersive Analytics

Virtual and augmented reality systems and applications have proven to be effective tools for visualization of large-scale data. In this paper, we present our results of two experiments carried out in a VR system aimed at visual analysis of graph visualization. We evaluated the performance of existing saliency models to predict salient scene regions during visual analysis that tackled both bottom-up and top-down visual attention aspects. Our findings support the fact that current visual saliency models are on one hand capable of detecting visual change in the VR scene, but on the other hand bottom-up saliency models do not predict well salient regions of a VR scene over time.