MMCTAgent: Multi-modal Critical Thinking Agent Framework for Complex Visual Reasoning

TL;DR

MMCTAgent is a multi-modal critical thinking framework that enhances complex visual reasoning by iterative analysis, decomposition, planning, and self-reflection, outperforming existing models on various benchmarks.

Contribution

Introduces MMCTAgent, a novel multi-modal critical thinking framework that improves reasoning in vision-language tasks through iterative analysis and self-evaluation.

Findings

Outperforms baseline MLLMs on multiple benchmarks

Enhances reasoning accuracy with critical thinking components

Demonstrates effectiveness in image and video understanding tasks

Abstract

Recent advancements in Multi-modal Large Language Models (MLLMs) have significantly improved their performance in tasks combining vision and language. However, challenges persist in detailed multi-modal understanding, comprehension of complex tasks, and reasoning over multi-modal information. This paper introduces MMCTAgent, a novel multi-modal critical thinking agent framework designed to address the inherent limitations of current MLLMs in complex visual reasoning tasks. Inspired by human cognitive processes and critical thinking, MMCTAgent iteratively analyzes multi-modal information, decomposes queries, plans strategies, and dynamically evolves its reasoning. Additionally, MMCTAgent incorporates critical thinking elements such as verification of final answers and self-reflection through a novel approach that defines a vision-based critic and identifies task-specific evaluation criteria, thereby enhancing its decision-making abilities. Through rigorous evaluations across various image and video understanding benchmarks, we demonstrate that MMCTAgent (with and without the critic) outperforms both foundational MLLMs and other tool-augmented pipelines.