Publication
Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning
Vision Large Language Models, Reasoning
Cite this work View paper
Citation
Di Zhang, Jingdi Lei, Junxian Li, Xunzhi Wang, Yujie Liu, Zonglin Yang, Jiatong Li, Weida Wang, Suorong Yang, Jianbo Wu, Peng Ye, Wanli Ouyang, Dongzhan Zhou. (2025). Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning. CVPR 2025.
About this work
Critic-V pairs a Reasoner with a separate Critic for multimodal reasoning. The Reasoner produces a response from visual and textual inputs, and the Critic provides natural-language feedback for revision.
The Critic is trained with Direct Preference Optimization on critiques ranked by a rule-based reward. The study evaluates reasoning accuracy and efficiency across eight benchmarks.