将视频理解难度分解为视觉感知与认知推理两轴，实现更高效的强化学习训练。
VideoCuRL: Video Curriculum Reinforcement Learning with Orthogonal Difficulty Decomposition
- 用光流和关键帧熵量化视觉复杂度，用校准惊喜度衡量认知难度，构建二维课程网格。
- 在VSI-Bench上推理能力提升2.5，在VideoMME上感知任务提升2.9。
- 适合需要高效视频后训练的场景，尤其适合资源受限环境部署。
强化学习（RL）对于赋予视频大模型复杂时空推理能力至关重要。然而，当前多数RL范式依赖随机数据打乱或基于标量难度指标的简单课程策略。我们认为，标量指标无法解耦视频理解中的两个正交挑战：视觉时间感知负荷与认知推理深度。为此，我们提出VideoCuRL，一种将难度分解为这两个维度的新框架。采用无需训练的高效代理：光流与关键帧熵用于视觉复杂度，校准惊喜度用于认知复杂度，将数据映射至二维课程网格。通过具备能力感知的对角波前策略，从基础对齐逐步推进到复杂推理。此外，引入动态稀疏KL与结构化重访机制，缓解奖励崩溃与灾难性遗忘问题。大量实验表明，VideoCuRL在推理任务（VSI-Bench +2.5）与感知任务（VideoMME +2.9）上均超越强基线。值得注意的是，VideoCuRL消除了生成式课程带来的高昂推理开销，提供了一种可扩展的鲁棒视频后训练方案。
原文摘要 · Abstract (English)
Reinforcement Learning (RL) is crucial for empowering VideoLLMs with complex spatiotemporal reasoning. However, current RL paradigms predominantly rely on random data shuffling or naive curriculum strategies based on scalar difficulty metrics. We argue that scalar metrics fail to disentangle two orthogonal challenges in video understanding: Visual Temporal Perception Load and Cognitive Reasoning Depth. To address this, we propose VideoCuRL, a novel framework that decomposes difficulty into these two axes. We employ efficient, training-free proxies, optical flow and keyframe entropy for visual complexity, Calibrated Surprisal for cognitive complexity, to map data onto a 2D curriculum grid. A competence aware Diagonal Wavefront strategy then schedules training from base alignment to complex reasoning. Furthermore, we introduce Dynamic Sparse KL and Structured Revisiting to stabilize training against reward collapse and catastrophic forgetting. Extensive experiments show that VideoCuRL surpasses strong RL baselines on reasoning (+2.5 on VSI-Bench) and perception (+2.9 on VideoMME) tasks. Notably, VideoCuRL eliminates the prohibitive inference overhead of generation-based curricula, offering a scalable solution for robust video post-training.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。