Look here! A parametric learning based approach to redirect visual attention

Across photography, marketing, and website design, being able to direct the\nviewer's attention is a powerful tool. Motivated by professional workflows, we\nintroduce an automatic method to make an image region more attention-capturing\nvia subtle image edits that maintain realism and fidelity to the original. From\nan input image and a user-provided mask, our GazeShiftNet model predicts a\ndistinct set of global parametric transformations to be applied to the\nforeground and background image regions separately. We present the results of\nquantitative and qualitative experiments that demonstrate improvements over\nprior state-of-the-art. In contrast to existing attention shifting algorithms,\nour global parametric approach better preserves image semantics and avoids\ntypical generative artifacts. Our edits enable inference at interactive rates\non any image size, and easily generalize to videos. Extensions of our model\nallow for multi-style edits and the ability to both increase and attenuate\nattention in an image region. Furthermore, users can customize the edited\nimages by dialing the edits up or down via interpolations in parameter space.\nThis paper presents a practical tool that can simplify future image editing\npipelines.\n