FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Models

• URL: http://arxiv.org/abs/2501.01986v2
• Date: Thu, 24 Jul 2025 18:44:26 GMT
• Title: FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Models
• Authors: Tianyu Fu, Tengxuan Liu, Qinghao Han, Guohao Dai, Shengen Yan, Huazhong Yang, Xuefei Ning, Yu Wang,
• Abstract summary: Existing token reduction methods primarily prune tokens based on importance metrics, such as cumulative attention scores.<n>We propose FrameFusion, a novel token reduction approach integrating similarity-based merging with importance-based pruning.<n>Experiments show that FrameFusion reduces visual tokens by 70%, achieving 1.6-3.6x end-to-end speedups, with an average performance impact of less than 3%.
• Score: 17.796102228521693
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: The increasing demand to process long and high-resolution videos significantly burdens Large Vision-Language Models (LVLMs) due to the enormous number of visual tokens. Existing token reduction methods primarily prune tokens based on importance metrics, such as cumulative attention scores. However, even important tokens may exhibit high redundancy caused by similarity among adjacent video frames and repetitive visual elements. To address this limitation, we propose FrameFusion, a novel token reduction approach integrating similarity-based merging with importance-based pruning. We conduct a thorough study on token similarity characteristics, revealing three key insights: (1) spatially corresponding visual tokens between adjacent frames have higher cosine similarities compared to other token pairs; (2) high token similarities prominently decrease in deeper model layers; and (3) token similarity rankings are highly consistent across different layers. Guided by these observations, FrameFusion computes token similarities exclusively between corresponding visual tokens from adjacent frames, applies token merging at initial successive layers followed by pruning in deeper layers, and adopts a cascaded merging strategy to further enhance efficiency. We evaluate FrameFusion comprehensively across six diverse LVLMs, ranging from 2B to 72B parameters, using five video benchmarks encompassing video retrieval, question-answering, and spatial-temporal understanding tasks. Experiments show that FrameFusion reduces visual tokens by 70%, achieving 1.6-3.6x end-to-end speedups, with an average performance impact of less than 3%. Our code is available at: https://github.com/thu-nics/FrameFusion.

Related papers

• REEF: Relevance-Aware and Efficient LLM Adapter for Video Understanding [2.309018557701645]
  Recent methods often compress memory banks to handle untemporal videos for video-level understanding. To this, we designed video to compress un videos on a large scale using visual tokens.
  arXiv  Detail & Related papers  (2025-04-07T20:36:34Z)
• QuoTA: Query-oriented Token Assignment via CoT Query Decouple for Long Video Comprehension [86.0749609778104]
  We propose QuoTA, an ante-hoc training-free modular that extends existing large video-language models. QuoTA strategically allocates frame-level importance scores based on query relevance. We decouple the query through Chain-of-Thoughts reasoning to facilitate more precise LVLM-based frame importance scoring.
  arXiv  Detail & Related papers  (2025-03-11T17:59:57Z)
• Token-Efficient Long Video Understanding for Multimodal LLMs [101.70681093383365]
  STORM is a novel architecture incorporating a dedicated temporal encoder between the image encoder and the Video-LLMs. We show that STORM achieves state-of-the-art results across various long video understanding benchmarks.
  arXiv  Detail & Related papers  (2025-03-06T06:17:38Z)
• ReTaKe: Reducing Temporal and Knowledge Redundancy for Long Video Understanding [55.320254859515714]
  We introduce a training-free method, $bfReTaKe$, to reduce both temporal visual redundancy and knowledge redundancy for long video understanding.<n> DPSelect identifies Videos with local maximum peak distance based on their visual features, which are closely aligned with human video perception.<n> PivotKV employs VideoBenchs as pivots and conducts KV-Cache compression for the non-text tokens with low attention scores.
  arXiv  Detail & Related papers  (2024-12-29T15:42:24Z)
• Importance-Based Token Merging for Efficient Image and Video Generation [41.94334394794811]
  We show that preserving high-information tokens during merging significantly improves sample quality.<n>We propose an importance-based token merging method that prioritizes the most critical tokens in computational resource allocation.
  arXiv  Detail & Related papers  (2024-11-23T02:01:49Z)
• Inference Optimal VLMs Need Only One Visual Token but Larger Models [54.01228554126122]
  Vision Language Models (VLMs) have demonstrated strong capabilities across various visual understanding and reasoning tasks. VLMs are often constrained by high latency during inference due to substantial compute required to process the large number of input tokens. We take some initial steps towards building approaches tailored for high token compression settings.
  arXiv  Detail & Related papers  (2024-11-05T18:54:21Z)
• Video Token Merging for Long-form Video Understanding [17.59960070514554]
  We propose a learnable video token merging algorithm that dynamically merges tokens based on their saliency. Our approach significantly reduces memory costs by 84% and boosts throughput by approximately 6.89 times compared to baseline algorithms.
  arXiv  Detail & Related papers  (2024-10-31T09:55:32Z)
• VideoLLM-MoD: Efficient Video-Language Streaming with Mixture-of-Depths Vision Computation [66.00245701441547]
  We introduce a novel approach to reduce vision compute by leveraging redundant vision tokens "skipping layers" rather than decreasing the number of vision tokens. Our method, VideoLLM-MoD, is inspired by mixture-of-depths LLMs and addresses the challenge of numerous vision tokens in long-term or streaming video.
  arXiv  Detail & Related papers  (2024-08-29T17:21:58Z)
• VoCo-LLaMA: Towards Vision Compression with Large Language Models [56.20788367278211]
  Vision-Language Models (VLMs) have achieved remarkable success in various multi-modal tasks, but they are often bottlenecked by the limited context window. We propose VoCo-LLaMA, the first approach to compress vision tokens using LLMs. Our method achieves minimal performance loss with a compression ratio of 576$times$, resulting in up to 94.8$%$ fewer FLOPs and 69.6$%$ acceleration in inference time.
  arXiv  Detail & Related papers  (2024-06-18T05:05:12Z)
• Accelerating Transformers with Spectrum-Preserving Token Merging [43.463808781808645]
  PiToMe prioritizes the preservation of informative tokens using an additional metric termed the energy score. Experimental findings demonstrate that PiToMe saved from 40-60% FLOPs of the base models.
  arXiv  Detail & Related papers  (2024-05-25T09:37:01Z)
• LLaVA-PruMerge: Adaptive Token Reduction for Efficient Large Multimodal Models [35.88374542519597]
  Large Multimodal Models (LMMs) have shown significant visual reasoning capabilities by connecting a visual encoder and a large language model. Recent LMMs incorporate more complex visual inputs, such as high-resolution images and videos, which further increases the number of visual tokens significantly. We propose PruMerge, a novel adaptive visual token reduction strategy that significantly reduces the number of visual tokens without compromising the performance of LMMs.
  arXiv  Detail & Related papers  (2024-03-22T17:59:52Z)
• AiluRus: A Scalable ViT Framework for Dense Prediction [95.1313839257891]
  Vision transformers (ViTs) have emerged as a prevalent architecture for vision tasks owing to their impressive performance. We propose to apply adaptive resolution for different regions in the image according to their importance. We evaluate our proposed method on three different datasets and observe promising performance.
  arXiv  Detail & Related papers  (2023-11-02T12:48:43Z)
• Unified Coarse-to-Fine Alignment for Video-Text Retrieval [71.85966033484597]
  We propose a Unified Coarse-to-fine Alignment model, dubbed UCoFiA. Our model captures the cross-modal similarity information at different granularity levels. We apply the Sinkhorn-Knopp algorithm to normalize the similarities of each level before summing them.
  arXiv  Detail & Related papers  (2023-09-18T19:04:37Z)
• Correspondence Matters for Video Referring Expression Comprehension [64.60046797561455]
  Video Referring Expression (REC) aims to localize the referent objects described in the sentence to visual regions in the video frames. Existing methods suffer from two problems: 1) inconsistent localization results across video frames; 2) confusion between the referent and contextual objects. We propose a novel Dual Correspondence Network (dubbed as DCNet) which explicitly enhances the dense associations in both the inter-frame and cross-modal manners.
  arXiv  Detail & Related papers  (2022-07-21T10:31:39Z)
• Parameterization of Cross-Token Relations with Relative Positional Encoding for Vision MLP [52.25478388220691]
  Vision multi-layer perceptrons (MLPs) have shown promising performance in computer vision tasks. They use token-mixing layers to capture cross-token interactions, as opposed to the multi-head self-attention mechanism used by Transformers. We propose a new positional spacial gating unit (PoSGU) to efficiently encode the cross-token relations for token mixing.
  arXiv  Detail & Related papers  (2022-07-15T04:18:06Z)
• CenterCLIP: Token Clustering for Efficient Text-Video Retrieval [67.21528544724546]
  In CLIP, the essential visual tokenization process, which produces discrete visual token sequences, generates many homogeneous tokens due to the redundancy nature of consecutive frames in videos. This significantly increases computation costs and hinders the deployment of video retrieval models in web applications. In this paper, we design a multi-segment token clustering algorithm to find the most representative tokens and drop the non-essential ones.
  arXiv  Detail & Related papers  (2022-05-02T12:02:09Z)
• Disentangled Representation Learning for Text-Video Retrieval [51.861423831566626]
  Cross-modality interaction is a critical component in Text-Video Retrieval (TVR) We study the interaction paradigm in depth, where we find that its computation can be split into two terms. We propose a disentangled framework to capture a sequential and hierarchical representation.
  arXiv  Detail & Related papers  (2022-03-14T13:55:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.