Fully Semantic Gap Recovery for End-to-End Image Captioning | IEEE Journals & Magazine | IEEE Xplore