SVSR: A Self-Verification and Self-Rectification Paradigm for Multimodal Reasoning

Zhe Qian, Nianbing Su, Zhonghua Wang +6

Current multimodal models often suffer from shallow reasoning, leading to errors caused by incomplete or inconsistent thought processes. To address this limitation, we propose Self…