arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.14236cs.CVcs.AIcs.LGcs.PF
收藏

双信号自适应KV缓存优化用于视觉-语言模型中长形式视频理解

Dual-Signal Adaptive KV-Cache Optimization for Long-Form Video Understanding in Vision-Language Models

International Institute of Information Technology(国际信息研究所)

Vishnu Sai, Dheeraj Sai, Srinath B, Girish Varma, Priyesh Shukla

AI总结:

Sali-Cache通过双信号自适应缓存优化,提升视觉-语言模型处理长形式视频的内存效率与准确性。

摘要:

视觉-语言模型(VLMs)在处理长形式视频内容时面临关键的记忆瓶颈,因为键值(KV)缓存随着序列长度线性增长。现有解决方案主要采用反应性驱逐策略,在丢弃标记之前计算完整的注意力矩阵,导致大量计算浪费。我们提出了Sali-Cache,一种新的前瞻性优化框架,通过主动内存管理实现双信号自适应缓存。通过整合基于光流分析的时序滤波器来检测帧间冗余,以及利用显著性检测的空域滤波器来识别视觉显著区域,Sali-Cache在进入计算昂贵的注意力操作之前智能地管理内存分配。在LLaVA 1.6架构上的实验评估表明,我们的方法在有效内存使用上实现了2.20倍的压缩率,同时在BLEU、ROUGE-L和Exact Match指标上保持100%的准确性。此外,在相同内存预算约束下,Sali-Cache在延长的时间跨度内保留了丰富的上下文特征,而不降低模型性能,使在消费级硬件上高效处理长形式视频内容成为可能。

英文摘要:

Vision-Language Models (VLMs) face a critical memory bottleneck when processing long-form video content due to the linear growth of the Key-Value (KV) cache with sequence length. Existing solutions predominantly employ reactive eviction strategies that compute full attention matrices before discarding tokens, resulting in substantial computational waste. We propose Sali-Cache, a novel a priori optimization framework that implements dual-signal adaptive caching through proactive memory management. By integrating a temporal filter based on optical flow analysis for detecting inter-frame redundancy and a spatial filter leveraging saliency detection for identifying visually significant regions, Sali-Cache intelligently manages memory allocation before entering computationally expensive attention operations. Experimental evaluation on the LLaVA 1.6 architecture demonstrates that our method achieves a 2.20x compression ratio in effective memory usage while maintaining 100% accuracy across BLEU, ROUGE-L, and Exact Match metrics. Furthermore, under identical memory budget constraints, Sali-Cache preserves context-rich features over extended temporal durations without degrading model performance, enabling efficient processing of long-form video content on consumer-grade hardware.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑