Gravar-mail: Humans predict the forest, not the trees: statistical learning of spatiotemporal structure in visual scenes