Совместная энд-ту-энд архитектура

AbstractImage-to-text tasks such as open-ended image captioning and controllable image description have received extensive attention for decades. Here we advance this line of work further, presenting Visual Spatial Description (VSD), …...