AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Mitigating Low-Quality Reasoning in MLLMs: Self-Driven Refined Multimodal CoT with Selective Thinking and Step-wise Visual Enhancement

Chongjun Tu, Peng Ye, Dongzhan Zhou, Tao Chen, Wanli Ouyang

AAAI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a self-driven refined multimodal chain-of-thought approach for MLLMs that selectively enables thinking on hard queries and strengthens visual grounding at each reasoning step to reduce overthinking and low-quality reasoning.

Problem

Multimodal chain-of-thought (MCoT) in MLLMs suffers from low-quality reasoning: overthinking on simple queries and inefficient use of visual information, wasting computation.

Approach

MLLMs inherently can distinguish simple from difficult queries; the method exploits this to selectively trigger thinking and applies step-wise visual enhancement during reasoning, self-driven without external supervision (details not stated in the abstract).

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add selective thinking (entropy-based direct answer vs reasoning) (ablation)SDR-MCoT · CoMT/MMStar/HallusionBenchavg token consumption: 212.91 -> 74.07Improved
add selective thinking (ablation)SDR-MCoT · HallusionBenchaccuracy: 63.76 -> 67.39Improved
add step-wise visual enhancement (ablation)SDR-MCoT · HallusionBenchaccuracy: 63.76 -> 67.76Improved
add step-wise visual enhancement (ablation)SDR-MCoT · token consumptiontokens per answer: 212.91 -> 216.57No clear effect
combine selective thinking and visual enhancement (ablation)SDR-MCoT · MMStaraccuracy: 53.62 -> 56.39Improved
apply zero-shot CoT to base MLLMQwen2-VL-7B · average over 8 benchmarksaverage accuracy: 59.00 -> 57.34Hurt
apply zero-shot CoT to base MLLMInternVL3-8B · average over 8 benchmarksaverage accuracy: 70.28 -> 65.90Hurt
apply SDR-MCoT instead of CoTQwen2-VL-7B · average over 8 benchmarksaverage accuracy: 57.34 -> 63.69Improved
apply SDR-MCoT instead of CoTInternVL3-8B · average over 8 benchmarksaverage accuracy: 65.90 -> 72.87Improved
apply SDR-MCoTLLaVA-1.5-13B · average over 8 benchmarksaverage accuracy: 46.40 -> 48.57Improved
apply visual-supplementation CoT (CCoT, DDCoT)Qwen2-VL-7B · average over 8 benchmarksaverage accuracy and tokens: CCoT 57.30 avg, 920.35 tokensHurt

Surprising findings

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy