Learning for Visual Synthesis and Transformation