FreqCache: Accelerating Embodied VLN Models with Adaptive Frequency-Guided Token Caching
FreqCache: 通过自适应频率引导的标记缓存加速具身VLN模型
机构 * School of Computer Science, Peking University(北京大学计算机科学学院) ; Beijing Academy of Artificial Intelligence, BAAI(北京人工智能研究院) ; School of Artificial Intelligence and Automation, Huazhong University of Science and Technology(华中科技大学人工智能与自动化学院) ; School of Computer Science, China University of Geosciences (Wuhan)(中国地质大学(武汉)计算机科学学院) ; College of Computer Science and Electronic Engineering, Hunan University(湖南大学计算机科学与电子工程学院) ; School of EECS, Peking University(北京大学电子信息技术学院)
AI总结 本文提出FreqCache框架,通过频率域方法优化VLN模型中的标记缓存,解决传统视觉方法在视角迁移、边缘信息和场景时变性上的不足,实验显示其在计算效率上有1.59倍的提升。