Cantor

Cantor is a multi-modal chain of thinking (CoT) framework that combines visual context acquisition with logical reasoning through a perceptual decision-making architecture to solve complex visual reasoning tasks. Cantor first acts as a decision generator, integrating visual input to analyze images and questions, ensuring closer alignment with real-world situations. In addition, Cantor leverages the advanced cognitive capabilities of large language models (MLLMs) to act as multi-faceted experts to derive higher-level information and enhance the CoT generation process. Cantor conducts extensive experiments on two complex visual reasoning datasets, demonstrating the effectiveness of the proposed framework, significantly improving multi-modal CoT performance without fine-tuning or ground-truth justification.

Cantor's target audience is professionals in the field of education and research, especially those researchers and educators who need to solve complex visual reasoning tasks. Cantor's multi-modal chain thinking framework can help them analyze images and questions more effectively, generate more accurate decisions and answers, thereby improving the quality of research and teaching.