Video-MTR: Reinforced Multi-Turn Reasoning for Long Video Understanding

Long-form video understanding remains a formidable challenge due to the complexity of modeling long-range temporal dependencies and multi-event narratives. Existing methods often rely on static reasoning or external Visual-Language Models (VLMs), resulting in high computational complexity and sub-optimal performance. In this paper, we propose Video-MTR, a reinforced multi-turn reasoning framework that operates solely through data-efficient, pure RL post-training. Video-MTR reformulates video understanding as a dynamic decision-making process, where the agent iteratively selects key segments conditioned on the evolving context of previously processed frames and the query. To ensure effective intermediate reasoning and training stability, we introduce a novel gated bi-level reward system, which synergizes trajectory-level rewards (answer correctness) with turn-level rewards (frame-query relevance). This mechanism eliminates the need for data-intensive supervised fine-tuning, thereby substantially reducing reliance on large-scale datasets. Remarkably, Video-MTR achieves competitive or superior performance using only $\sim$8K training samples, compared to existing approaches that require 257K to 4.4M examples. Extensive experiments on benchmarks including VideoMME, MLVU, LongVideoBench, LVBench, and EgoSchema demonstrate that Video-MTR surpasses state-of-the-art methods in both accuracy and efficiency. Code is available at https://github.com/Xyuan13/Video-MTR.

Long videos are difficult for AI systems to understand because important moments may appear only briefly and far apart in time. Existing video-language models often look at a fixed set of sampled frames, which can miss the key evidence needed to answer a question. We introduce Video-MTR, a method that teaches a model to inspect a long video in multiple steps. The model first forms a broad view of the video, then actively selects more relevant video segments, and finally uses the gathered evidence to answer the question. To train this behavior efficiently, we design a reward system that encourages both correct final answers and useful intermediate evidence selection. This allows Video-MTR to learn effective long-video reasoning with only about 8K training examples, far fewer than many existing approaches. Our results show that this multi-step strategy improves accuracy and efficiency across several long-video understanding benchmarks.