Lune

ACL Findings2025

AdaReTaKe: Adaptive Redundancy Reduction to Perceive Longer for Video-language Understanding

Xiao Wang, Qingyi Si, Shiyu Zhu, Jianlong Wu, Li Cao, Liqiang Nie

Abstract

Multimodal Large Language Models (MLLMs) have revolutionized video understanding, yet are still limited by context length when processing long videos.

Recent methods compress videos by leveraging visual redundancy uniformly, yielding promising results.

Nevertheless, our quantitative analysis shows that redundancy varies significantly across time and model layers, necessitating a more flexible compression strategy.

We propose AdaRETAKE, a training-free method that flexibly reduces visual redundancy by allocating compression ratios among time and layers with theoretical guarantees. AdaRETAKE can be seamlessly integrated into existing MLLMs as a plugand-play solution, extending their processing capacity from 256 to 2048 frames while preserving critical information. Experiments on VideoMME, MLVU, LongVideoBench, and LVBench datasets demonstrate that AdaRETAKE outperforms existing methods by 2.3% and 2.8% for 7B and 72B models, respectively, with even greater improvements of 5.9% and 6.0% on the longest LVBench. Our code is available at https://github.com/ SCZwangxiao/video-FlexReduc.git.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

15 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding cites LVBench: An Extreme Long Video Understanding Benchmark
  • FlexSelect: Flexible Token Selection for Efficient Long Video Understanding cites LVBench: An Extreme Long Video Understanding Benchmark
  • VideoARM: Agentic Reasoning over Hierarchical Memory for Long-Form Video Understanding cites Deep Video Discovery: Agentic Search with Tool Use for Long-form Video Understanding
  • VideoARM: Agentic Reasoning over Hierarchical Memory for Long-Form Video Understanding cites VideoLucy: Deep Memory Backtracking for Long Video Understanding
  • LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context Inference cites Efficient Streaming Language Models with Attention Sinks
  • LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context Inference cites Token Merging: Your ViT But Faster
  • VideoLucy: Deep Memory Backtracking for Long Video Understanding cites LVBench: An Extreme Long Video Understanding Benchmark
  • Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language Models cites Token Merging: Your ViT But Faster
  • Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference cites Efficient Streaming Language Models with Attention Sinks
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers15

  1. Deep Video Discovery: Agentic Search with Tool Use for Long-form Video UnderstandingXiaoyi Zhang, Zhaoyang Jia, Zongyu Guo, Jiahao Li et al.NeurIPS 2025 · 95 citations
  2. LVBench: An Extreme Long Video Understanding BenchmarkWeihan Wang, Zehai He, Wenyi Hong, Yean Cheng et al.ICCV 2025 · 28 citations
  3. VideoARM: Agentic Reasoning over Hierarchical Memory for Long-Form Video UnderstandingYufei Yin, Qianke Meng, Minghao Chen, Jiajun Ding et al.CVPR 2026 · 24 citations
  4. VideoLucy: Deep Memory Backtracking for Long Video UnderstandingJialong Zuo, Yongtai Deng, Lingdong Kong, Jingkang Yang et al.NeurIPS 2025 · 23 citations
  5. FlexSelect: Flexible Token Selection for Efficient Long Video UnderstandingYunzhu Zhang, Yu Lu, Tianyi Wang, Fengyun Rao et al.NeurIPS 2025 · 22 citations

Builds on16

  1. Efficient Streaming Language Models with Attention SinksGuangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han et al.ICLR 2024 · 1,714 citations
  2. Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM InferenceYuan Feng, Junlin Lv, Yukun Cao, Xike Xie et al.NeurIPS 2025 · 256 citations
  3. Fit and Prune: Fast and Training-free Visual Token Pruning for Multi-modal Large Language ModelsWeihao Ye, Qiong Wu, Wenhao Lin, Yiyi ZhouAAAI 2025 · 99 citations
  4. Token Merging: Your ViT But FasterDaniel Bolya, Cheng-Yang Fu, Xiaoliang Dai, Peizhao Zhang et al.ICLR 2023 · 62 citations
  5. LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context InferenceZhongwei Wan, Ziang Wu, Che Liu, Jinfa Huang et al.EMNLP Findings 2024 · 21 citations

Related papers

  1. LongVU: Spatiotemporal Adaptive Compression for Long Video-Language UnderstandingXiaoqian Shen, Yunyang Xiong, Changsheng Zhao, Lemeng Wu et al.ICML 2025
  2. EAKV: An Entropy-Driven Adaptive KV Compression Framework for Long Video UnderstandingHengrui Hu, Jingyu Li, Juntao Liang, Guanyu Chen et al.ICML 2026
  3. Free-Moref: Instantly Multiplexing Context Perception Capabilities of Video-Mllms Within Single InferenceKuo Wang, Quanlong Zheng, Junlin Xie, Yanhao Zhang et al.ICCV 2025
  4. AIM: Adaptive Inference of Multi-Modal LLMs via Token Merging and PruningYiwu Zhong, Zhuoming Liu, Yin Li, Liwei WangICCV 2025 · 1 citation
  5. METok: Multi-Stage Event-based Token Compression for Efficient Long Video UnderstandingMengyue Wang, Shuo Chen, Kristian Kersting, Volker Tresp et al.EMNLP 2025 · 5 citations
Dusk over the sea between two cliffs drawn in fine vertical lines