Region-Object Relation-Aware Dense Captioning via Transformer | IEEE Journals & Magazine | IEEE Xplore