STATE: Learning structure and texture representations for novel view synthesis