RA3: A Human-in-the-loop Framework for Interpreting and Improving Image Captioning with Relation-Aware Attribution Analysis.

Lei Chai, Lu Qi, Hailong Sun 0001, Jingzheng Li