Reinforcing Structured Chain-of-Thought for Video Understanding

TL;DR

This paper proposes SDRL, a single-stage reinforcement learning framework with structured reasoning for video understanding, eliminating the need for costly annotations and improving performance on VideoQA tasks.

Contribution

It introduces a novel single-stage RL approach with structured CoT format and self-supervised mechanisms, enhancing reasoning and generalization in MLLMs for video understanding.

Findings

Achieves state-of-the-art results on seven VideoQA datasets.

Effectively balances factual grounding and reasoning diversity.

Eliminates the need for supervised fine-tuning annotations.

Abstract

Multi-modal Large Language Models (MLLMs) show promise in video understanding. However, their reasoning often suffers from thinking drift and weak temporal comprehension, even when enhanced by Reinforcement Learning (RL) techniques like Group Relative Policy Optimization (GRPO). Moreover, existing RL methods usually depend on Supervised Fine-Tuning (SFT), which requires costly Chain-of-Thought (CoT) annotation and multi-stage training, and enforces fixed reasoning paths, limiting MLLMs' ability to generalize and potentially inducing bias. To overcome these limitations, we introduce Summary-Driven Reinforcement Learning (SDRL), a novel single-stage RL framework that obviates the need for SFT by utilizing a Structured CoT format: Summarize -> Think -> Answer. SDRL introduces two self-supervised mechanisms integrated into the GRPO objective: 1) Consistency of Vision Knowledge (CVK) enforces factual grounding by reducing KL divergence among generated summaries; and 2) Dynamic Variety of Reasoning (DVR) promotes exploration by dynamically modulating thinking diversity based on group accuracy. This novel integration effectively balances alignment and exploration, supervising both the final answer and the reasoning process. Our method achieves state-of-the-art performance on seven public VideoQA datasets.