Reinforcing Video Reasoning with Focused Thinking

TL;DR

This paper introduces TW-GRPO, a reinforcement learning framework that improves video reasoning by focusing on salient information and utilizing dense rewards, leading to state-of-the-art results on multiple benchmarks.

Contribution

The paper proposes TW-GRPO, which enhances visual reasoning with token weighting, multi-choice rewards, and data augmentation, addressing key limitations of previous methods.

Findings

Achieves 50.4% accuracy on CLEVRER, outperforming previous models.

Improves MMVU accuracy to 65.8%.

Demonstrates effectiveness of focused reasoning and dense rewards.

Abstract

Recent advancements in reinforcement learning, particularly through Group Relative Policy Optimization (GRPO), have significantly improved multimodal large language models for complex reasoning tasks. However, two critical limitations persist: 1) they often produce unfocused, verbose reasoning chains that obscure salient spatiotemporal cues and 2) binary rewarding fails to account for partially correct answers, resulting in high reward variance and inefficient learning. In this paper, we propose TW-GRPO, a novel framework that enhances visual reasoning with focused thinking and dense reward granularity. Specifically, we employs a token weighting mechanism that prioritizes tokens with high informational density (estimated by intra-group information entropy), suppressing redundant tokens like generic reasoning prefixes. Furthermore, we reformulate RL training by shifting from single-choice to multi-choice QA tasks, where soft rewards enable finer-grained gradient estimation by distinguishing partial correctness. Additionally, we propose question-answer inversion, a data augmentation strategy to generate diverse multi-choice samples from existing benchmarks. Experiments demonstrate state-of-the-art performance on several video reasoning and general understanding benchmarks. Notably, TW-GRPO achieves 50.4\% accuracy on CLEVRER (18.8\% improvement over Video-R1) and 65.8\% on MMVU. Our codes are available at \href{https://github.com/longmalongma/TW-GRPO}.