Video-Thinker: Sparking "Thinking with Videos" via Reinforcement Learning

2025-10-27

AI Summary

Existing video reasoning methods lack an autonomous reasoning mechanism analogous to the “chain-of-thought” (CoT) paradigm in image understanding. This paper introduces Video-Thinker, the first end-to-end video reasoning framework that extends the image CoT paradigm from multimodal large language models (MLLMs) to video, without requiring external tools. Video-Thinker employs reinforcement learning to enable the model to autonomously invoke its built-in spatiotemporal localization and descriptive capabilities during inference, generating temporally coherent reasoning traces. It adopts joint training via supervised fine-tuning and grouped relative policy optimization. Furthermore, we construct Video-Thinker-10K—the first video reasoning dataset explicitly designed for autonomous tool invocation—containing 10K high-quality samples. On multiple video reasoning benchmarks, our 7B-parameter model substantially outperforms strong baselines such as Video-R1 and achieves state-of-the-art performance among models of comparable scale.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Recent advances in image reasoning methods, particularly "Thinking with Images", have demonstrated remarkable success in Multimodal Large Language Models (MLLMs); however, this dynamic reasoning paradigm has not yet been extended to video reasoning tasks. In this paper, we propose Video-Thinker, which empowers MLLMs to think with videos by autonomously leveraging their intrinsic "grounding" and "captioning" capabilities to generate reasoning clues throughout the inference process. To spark this capability, we construct Video-Thinker-10K, a curated dataset featuring autonomous tool usage within chain-of-thought reasoning sequences. Our training strategy begins with Supervised Fine-Tuning (SFT) to learn the reasoning format, followed by Group Relative Policy Optimization (GRPO) to strengthen this reasoning capability. Through this approach, Video-Thinker enables MLLMs to autonomously navigate grounding and captioning tasks for video reasoning, eliminating the need for constructing and calling external tools. Extensive experiments demonstrate that Video-Thinker achieves significant performance gains on both in-domain tasks and challenging out-of-domain video reasoning benchmarks, including Video-Holmes, CG-Bench-Reasoning, and VRBench. Our Video-Thinker-7B substantially outperforms existing baselines such as Video-R1 and establishes state-of-the-art performance among 7B-sized MLLMs.

Problem

• Extends dynamic reasoning paradigm to video tasks
• Enables autonomous grounding and captioning for video reasoning
• Eliminates need for external tools in video understanding

Innovation

• Uses reinforcement learning to enhance video reasoning
• Autonomously leverages grounding and captioning capabilities
• Employs supervised fine-tuning and group policy optimization