Slow-Fast Architecture for Video Multi-Modal Large Language Models

• URL: http://arxiv.org/abs/2504.01328v1
• Date: Wed, 02 Apr 2025 03:24:58 GMT
• Title: Slow-Fast Architecture for Video Multi-Modal Large Language Models
• Authors: Min Shi, Shihao Wang, Chieh-Yun Chen, Jitesh Jain, Kai Wang, Junjun Xiong, Guilin Liu, Zhiding Yu, Humphrey Shi,
• Abstract summary: Existing methods compress video representations using predefined rules before feeding them into the multi-modal large language model.<n>We propose a novel slow-fast architecture that naturally circumvents this trade-off, enabling the use of more input frames while preserving spatial details.<n>Our model significantly outperforms self-attention-only baselines, extending the input capacity from 16 to 128 frames with just a 3% increase in computation.
• Score: 42.3957835391319
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Balancing temporal resolution and spatial detail under limited compute budget remains a key challenge for video-based multi-modal large language models (MLLMs). Existing methods typically compress video representations using predefined rules before feeding them into the LLM, resulting in irreversible information loss and often ignoring input instructions. To address this, we propose a novel slow-fast architecture that naturally circumvents this trade-off, enabling the use of more input frames while preserving spatial details. Inspired by how humans first skim a video before focusing on relevant parts, our slow-fast design employs a dual-token strategy: 1) "fast" visual tokens -- a compact set of compressed video features -- are fed into the LLM alongside text embeddings to provide a quick overview; 2) "slow" visual tokens -- uncompressed video features -- are cross-attended by text embeddings through specially designed hybrid decoder layers, enabling instruction-aware extraction of relevant visual details with linear complexity. We conduct systematic exploration to optimize both the overall architecture and key components. Experiments show that our model significantly outperforms self-attention-only baselines, extending the input capacity from 16 to 128 frames with just a 3% increase in computation, and achieving a 16% average performance improvement across five video understanding benchmarks. Our 7B model achieves state-of-the-art performance among models of similar size. Furthermore, our slow-fast architecture is a plug-and-play design that can be integrated into other video MLLMs to improve efficiency and scalability.

Related papers

• FiLA-Video: Spatio-Temporal Compression for Fine-Grained Long Video Understanding [17.71123451197036]
  complexity of video data and contextual processing limitations still hinder long-video comprehension. We propose FiLA-Video, a novel framework that integrates multiple frames into a single representation. FiLA-Video achieves superior efficiency and accuracy in long-video comprehension compared to existing methods.
  arXiv  Detail & Related papers  (2025-04-29T03:09:46Z)
• FastVID: Dynamic Density Pruning for Fast Video Large Language Models [38.267065642416554]
  We propose Dynamic Density Pruning for Fast Video LLMs termed FastVID. FastVID partitions videos into temporally ordered segments to preserve temporal structure and applies a density-based token pruning strategy to maintain essential visual information. Our method significantly reduces computational overhead while maintaining temporal and visual integrity.
  arXiv  Detail & Related papers  (2025-03-14T08:33:08Z)
• 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.<n>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)
• Video-Panda: Parameter-efficient Alignment for Encoder-free Video-Language Models [26.866184981409607]
  We present an efficient encoder-free approach for video-language understanding that achieves competitive performance while significantly reducing computational overhead.<n>Our method introduces a novel Spatio-Temporal Alignment Block (STAB) that directly processes video inputs without requiring pre-trained encoders.<n>Our model achieves comparable or superior performance to encoder-based approaches for open-ended video question answering on standard benchmarks.
  arXiv  Detail & Related papers  (2024-12-24T18:59:56Z)
• LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding [65.46303012350207]
  LongVU is an adaptive compression mechanism that reduces the number of video tokens while preserving visual details of long videos. We leverage DINOv2 features to remove redundant frames that exhibit high similarity. We perform spatial token reduction across frames based on their temporal dependencies.
  arXiv  Detail & Related papers  (2024-10-22T21:21:37Z)
• Oryx MLLM: On-Demand Spatial-Temporal Understanding at Arbitrary Resolution [90.31313348540607]
  We propose Oryx, a unified multimodal architecture for the spatial-temporal understanding of images, videos, and 3D scenes. Oryx offers an on-demand solution to seamlessly and efficiently process visual inputs with arbitrary spatial sizes and temporal lengths. Design features enable Oryx to accommodate extremely long visual contexts, such as videos, with lower resolution and high compression.
  arXiv  Detail & Related papers  (2024-09-19T17:59:51Z)
• SlowFast-LLaVA: A Strong Training-Free Baseline for Video Large Language Models [51.712700398020075]
  We propose a training-free video large language model (LLM) that can jointly capture detailed spatial semantics and long-range temporal context. This is realized by using a two-stream SlowFast design of inputs for Video LLMs to aggregate features from sampled frames in an effective way. Experimental results show that SF-LLaVA outperforms existing training-free methods on a wide range of video tasks.
  arXiv  Detail & Related papers  (2024-07-22T17:58:04Z)
• A Simple Recipe for Contrastively Pre-training Video-First Encoders Beyond 16 Frames [57.758863967770594]
  We build on the common paradigm of transferring large-scale, image--text models to video via shallow temporal fusion.<n>We expose two limitations to the approach: (1) decreased spatial capabilities, likely due to poor video--language alignment in standard video datasets, and (2) higher memory consumption, bottlenecking the number of frames that can be processed.
  arXiv  Detail & Related papers  (2023-12-12T16:10:19Z)

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.