Explainable Knowledge reasoning via thought chains for knowledge-based visual question answering

Chen Qiu, Zhiqiang Xie, Maofu Liu, Huijun Hu

Abstract: Highlights•We investigate the role of multimodal CoT distilled knowledge from LLMs in KBVQA tasks.•We present a Multimodal Knowledge Reasoning via CoT (MuKCoT) model for reasoning and explanation.•Performance of MuKCoT fosters the higher-quality explanations for the KBVQA tasks.