Generating Synthetic Images for Visual Attention Modeling