style information can be extracted from almost all layers, but don’t you think layers closer to input may have more specific details of your content while the later layers represent more general/abstract features of the content?