Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

DyCo-RL: Dynamic Cross-Modal Coordination for Visual Reasoning

About

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a leading paradigm for enhancing visual reasoning in Multimodal Large Language Models (MLLMs). However, existing RLVR methods optimize primarily for the reasoning outcome, fundamentally overlooking the fine-grained cross-modal coordination required during the generation process. Through token-level analyses and controlled interventions, we reveal that during Chain-of-Thought (CoT) reasoning, MLLMs frequently fail to dynamically alternate between extracting visual evidence and synthesizing textual context-a coordination breakdown that is causally linked to reasoning failures. Motivated by these findings, we propose DyCo-RL, which integrates dynamic cross-modal coordination into RLVR optimization. Specifically, DyCo-RL uses the Fisher-Rao geodesic distance to measure within-modality attention shifts, assigning tokens to either visually-oriented or text-oriented functional roles. It then evaluates the alignment between a token's actual attention allocation and its assigned role, leveraging this score for alignment-guided advantage reweighting during policy optimization. Extensive experiments demonstrate that the algorithm-agnostic DyCo-RL, when applied to Qwen2.5-VL-3B/7B, consistently improves four representative RLVR algorithms across seven benchmarks spanning visual-centric and mathematical reasoning.

Hangui Lin, Yan Shu, Zhengyang Liang, Chi Liu, Xiangrui Liu, Minghao Qin, Teng Long, Zheng Liu, Nicu Sebe• 2026

Related benchmarks

TaskDatasetResultRank
Visual Question AnsweringRealworldQA
Accuracy49.5
327
Mathematical ReasoningWeMath
Accuracy40.7
317
Mathematical ReasoningMathVerse--
266
Visual Question AnsweringA-OKVQA
Acc82.7
240
Visual Question AnsweringMMStar
Accuracy53.2
151
Visual Question AnsweringChartQA (test)
Accuracy83.2
95
Visual Question AnsweringSEED-Bench Image
Accuracy69
80
Visual ReasoningMMBench--
48
Visual ReasoningHallusionBench
Accuracy73.2
43
Mathematical ReasoningMathVision
Top-1 Accuracy29.3
27
Showing 10 of 12 rows

Other info

GitHub

Follow for update