仅用后训练提升流式视频理解，性能逼近大模型。
StreamOPD: A Post-Training Recipe with Spatio-Temporal Cue Gating for Streaming Video Understanding
- 提出StreamOPD方法，结合思维模式教学与指令部署。
- 在StreamingBench上从77.9%提升至83.9%，接近9B教师模型。
- 无需额外记忆或压缩，适合开源流式视频研究参考。
流式视频理解要求对不断展开的视频前缀做出即时响应。现有系统依赖推理时的记忆、检索与压缩，但无需训练的滑动窗口基线已能与之媲美。因此，本文固定无记忆的最近窗口协议，探索仅靠后训练能走多远。强化学习因验证奖励机制不匹配，易导致长篇思考生成；而在线策略蒸馏（OPD）虽提供密集的教师监督，却仅在双模型均以思考模式训练时稳定。由此提出StreamOPD：结合可验证流式视频数据、思考模式下的OPD及指令模式部署。该方法将StreamingBench准确率从77.9%提升至83.9%，仅差0.3个百分点于9B教师模型；在排除幻觉检测子任务的OVO-Bench上提升9.1点，且推理方式不变。进一步引入时空线索门控（ST-CueGate），通过聚合线索与非线索的教师似然比生成相对响应分数以重加权OPD，使OVO-Bench（不含HLD）达71.9%，Video-MME达64.9%，是唯一在四个基准上均优于基础模型的变体。若将教师替换为学生初始策略的冻结副本（即自蒸馏），仍保持大部分增益，并使HLD提升至57.0%，超过未训练学生与9B教师，表明回避损失并非该方案本质。本文提供透明可复现的开源流式视频研究基准。
原文摘要 · Abstract (English)
Streaming video understanding demands direct responses from the causally observed prefix of an unfolding video. Existing systems add inference-time memory, retrieval, and compression, yet a training-free sliding-window baseline already matches them. We therefore fix a memory-free recent-window protocol and ask how far post-training alone can go. Reinforcement learning with verifiable rewards fits this regime poorly, encouraging long ``think-then-answer'' generations, while on-policy distillation (OPD) supplies dense token-level teacher supervision on student trajectories but is stable only when both models train in thinking mode. These observations lead to \textsc{StreamOPD}, a recipe combining verifiable streaming-video data, thinking-mode OPD, and instruct-mode deployment. It raises StreamingBench from $77.9\%$ to $83.9\%$---within $0.3$ points of the 9B teacher---and improves OVO-Bench excluding its hallucination-detection subtask (HLD) by $9.1$ points under unchanged inference. As a teacher-privilege extension, \emph{Spatio-Temporal CueGate (ST-CueGate)} aggregates cue-versus-no-cue teacher likelihood ratios into a group-relative response score that reweights OPD. It reaches $71.9\%$ on OVO-Bench (excluding HLD) and $64.9\%$ on Video-MME, and is the only variant that stays above the base model on all four benchmarks. Replacing the teacher with a frozen copy of the student's initial policy---on-policy self-distillation---retains most of these gains and lifts HLD to $57.0\%$, above both the untrained student and the 9B teacher, so abstention loss is not intrinsic to the recipe. We provide a transparent and reproducible reference for open-source streaming-video research.
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。