ChainV: Atomic Visual Hints Make Multimodal Reasoning Shorter and Better

2025-11-21

AI Summary

Existing multimodal reasoning models suffer from redundant self-reflection and excessively long reasoning chains; moreover, mainstream training-free Chain-of-Thought (CoT) compression methods rely on static visual references, limiting adaptability to dynamic visual reasoning tasks. This paper proposes ChainV, a fine-tuning-free framework for dynamic visual prompt enhancement. Its core innovations include: (1) automatic selection of atomic-level visual prompts based on mean attention intensity; (2) adaptive control of reasoning depth via consistency evaluation and a Bernoulli stochastic process; and (3) integration of coarse-grained block filtering, attention-driven fine-grained visual extraction, reliability assessment, and dynamic thought injection. Evaluated on the MathVista benchmark, ChainV achieves a 2.3% accuracy gain, reduces inference latency by 51.4%, and decreases output token count by 24.5%, significantly improving both efficiency and accuracy in vision-dependent mathematical reasoning.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Recent advances in multimodal reasoning models have demonstrated impressive capabilities across text and vision. However, even leading models exhibit redundant self-reflection when generating lengthy reasoning chains. While training-free CoT compression methods have emerged in the LLMs domain, they rely on static visual references and thus provide limited gains for multimodal reasoning. Therefore, we propose ChainV, a framework that dynamically integrates visual hints into the reasoning process, thereby making multimodal reasoning shorter and better. Specifically, ChainV first performs a coarse visual patch selection based on the previous reasoning step, then refines it by identifying the most representative atomic visual hint according to the averaged attention intensity. Additionally, ChainV introduces a consistency-based evaluation mechanism to assess the reliability of the chosen hint, guiding the model to adaptively adjust its level of self-reflection. Eventually, the pixel coordinates of the selected visual hint and its reliability are incorporated into thinking with a Bernoulli stochastic process. Experiments indicate that our method significantly improves reasoning accuracy and efficiency, especially on math-intensive benchmarks where visual hints are crucial for multi-step symbolic reasoning. For example, ChainV achieves $2.3%$ improvement on the MathVista within MIMO-VL-RL, while reducing inference latency by $51.4%$ and shortening output token length by $24.5%$.

Problem

• Reduces redundant self-reflection in multimodal reasoning models
• Dynamically integrates atomic visual hints into reasoning chains
• Improves accuracy and efficiency on math-intensive visual benchmarks

Innovation

• Dynamically integrates visual hints into reasoning process
• Refines visual patch selection using attention intensity
• Uses Bernoulli process to incorporate hint reliability