AdaSpark: Adaptive Sparsity for Efficient Long-Video Understanding

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Handong, Liu, Zikang, Guo, Longteng, Yue, Tongtian, Tang, Yepeng, Zhu, Xinxin, Zheng, Chuanyang, Wang, Ziming, Wang, Zhibin, Song, Jun, Yu, Cheng, Zheng, Bo, Liu, Jing
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915990435004416
author Li, Handong
Liu, Zikang
Guo, Longteng
Yue, Tongtian
Tang, Yepeng
Zhu, Xinxin
Zheng, Chuanyang
Wang, Ziming
Wang, Zhibin
Song, Jun
Yu, Cheng
Zheng, Bo
Liu, Jing
author_facet Li, Handong
Liu, Zikang
Guo, Longteng
Yue, Tongtian
Tang, Yepeng
Zhu, Xinxin
Zheng, Chuanyang
Wang, Ziming
Wang, Zhibin
Song, Jun
Yu, Cheng
Zheng, Bo
Liu, Jing
contents Processing long-form videos with Video Large Language Models (Video-LLMs) is computationally prohibitive. Current efficiency methods often compromise fine-grained perception through irreversible information disposal or inhibit long-range temporal modeling via rigid, predefined sparse patterns. This paper introduces AdaSpark, an adaptive sparsity framework designed to address these limitations. AdaSpark first partitions video inputs into 3D spatio-temporal cubes. It then employs two co-designed, context-aware components: (1) Adaptive Cube-Selective Attention (AdaS-Attn), which adaptively selects a subset of relevant video cubes to attend for each query token, and (2) Adaptive Token-Selective FFN (AdaS-FFN), which selectively processes only the most salient tokens within each cube. An entropy-based (Top-p) selection mechanism adaptively allocates computational resources based on input complexity. Experiments demonstrate that AdaSpark significantly reduces computational load by up to 57% FLOPs while maintaining comparable performance to dense models and preserving fine-grained, long-range dependencies, as validated on challenging hour-scale video benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2604_08077
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle AdaSpark: Adaptive Sparsity for Efficient Long-Video Understanding
Li, Handong
Liu, Zikang
Guo, Longteng
Yue, Tongtian
Tang, Yepeng
Zhu, Xinxin
Zheng, Chuanyang
Wang, Ziming
Wang, Zhibin
Song, Jun
Yu, Cheng
Zheng, Bo
Liu, Jing
Computer Vision and Pattern Recognition
Processing long-form videos with Video Large Language Models (Video-LLMs) is computationally prohibitive. Current efficiency methods often compromise fine-grained perception through irreversible information disposal or inhibit long-range temporal modeling via rigid, predefined sparse patterns. This paper introduces AdaSpark, an adaptive sparsity framework designed to address these limitations. AdaSpark first partitions video inputs into 3D spatio-temporal cubes. It then employs two co-designed, context-aware components: (1) Adaptive Cube-Selective Attention (AdaS-Attn), which adaptively selects a subset of relevant video cubes to attend for each query token, and (2) Adaptive Token-Selective FFN (AdaS-FFN), which selectively processes only the most salient tokens within each cube. An entropy-based (Top-p) selection mechanism adaptively allocates computational resources based on input complexity. Experiments demonstrate that AdaSpark significantly reduces computational load by up to 57% FLOPs while maintaining comparable performance to dense models and preserving fine-grained, long-range dependencies, as validated on challenging hour-scale video benchmarks.
title AdaSpark: Adaptive Sparsity for Efficient Long-Video Understanding
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.08077