通过感知视频曲率实现高效压缩，提升长视频理解模型推理速度。
V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models
- 将令牌压缩视为轨迹近似问题，按语义转折点分配帧内令牌预算。
- 在保持原始坐标前提下，保留高熵视觉信息，避免位置错位。
- 无需训练、即插即用，显著降低内存与延迟，性能损失小于1.4%。
视频大语言模型（VideoLLMs）在视频理解任务中表现出强大能力，但长上下文推理仍受预填充阶段大量冗余视觉令牌的制约。本文在严格资源预算下重新审视视频令牌压缩问题，识别出关键瓶颈：时空信息覆盖不足。现有方法常因粗粒度帧级分配或场景分割导致覆盖不连续，且在MRoPE风格离散绑定下，令牌合并会进一步破坏时空对齐。为此，我们提出V-CAST（Video Curvature-Aware Spatio-Temporal Pruning），一种无需训练、即插即用的长视频推理剪枝策略。V-CAST将令牌压缩建模为轨迹逼近问题，引入曲率引导的时间分配模块，将每帧令牌预算动态分配至语义转折点与事件边界。同时采用双锚点空间选择机制，在无需注意力干预的情况下保留高熵视觉证据，并保持保留令牌的原始坐标以维持位置一致性。在多个不同架构与规模的VideoLLM上进行的大量实验表明，V-CAST达到原模型98.6%的性能，平均优于次优方法+1.1%，并将峰值内存和总延迟分别降低至Qwen3-VL-8B-Instruct的86.7%和86.4%。
原文摘要 · Abstract (English)
Video large language models (VideoLLMs) show strong capability in video understanding, yet long-context inference is still dominated by massive redundant visual tokens in the prefill stage. We revisit token compression for VideoLLMs under a tight budget and identify a key bottleneck, namely insufficient spatio-temporal information coverage. Existing methods often introduce discontinuous coverage through coarse per-frame allocation or scene segmentation, and token merging can further misalign spatio-temporal coordinates under MRoPE-style discrete (t,h,w) bindings. To address these issues, we propose V-CAST (Video Curvature-Aware Spatio-Temporal Pruning), a training-free, plug-and-play pruning policy for long-context video inference. V-CAST casts token compression as a trajectory approximation problem and introduces a curvature-guided temporal allocation module that routes per-frame token budgets to semantic turns and event boundaries. It further adopts a dual-anchor spatial selection mechanism that preserves high-entropy visual evidence without attention intervention, while keeping retained tokens at their original coordinates to maintain positional alignment. Extensive experiments across multiple VideoLLMs of different architectures and scales demonstrate that V-CAST achieves 98.6% of the original performance, outperforms the second-best method by +1.1% on average, and reduces peak memory and total latency to 86.7% and 86.4% of vanilla Qwen3-VL-8B-Instruct.
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。