CooKie: commonsense knowledge-guided mixture-of-experts framework for fine-grained visual question answering

Chao Wang, Jianming Yang, Yang Zhou, Xiaodong Yue

Abstract: Highlights•Our novel framework makes fine-grained visual question answering tasks more grounded.•We annotate more instances to do further analysis of our method.•Our method achieves SOTA performance on V* Bench, with 4% more accuracy.