Longteng Guo, Jing Liu, Jinhui Tang +3
The paper proposes a method for image captioning that aligns words with visual semantic units—objects, attributes, and object interactions—using semantic and geometry graphs proces…