Skip to content
AI News HubLIVE
Original source2 min read

OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs

Summary

OmniMem is a memory-efficient streaming framework for audio-visual LLMs that uses modality-aware memory allocation to separately manage visual and audio contexts, and perturbation-aware memory selection to preserve informative, non-redundant KV states. On VideoMME Long, LVBench, and LVOmniBench, it improves over strong training-free compression baselines by 2-4% absolute accuracy under the same memory budgets, with an additional 1-2% gain after fine-tuning.

SourcearXiv AIAuthor: Guangzhi Sun, Yixuan Li, Yudong Yang, Chao Zhang
OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[2606.07577] OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs

[Submitted on 26 May 2026]

Title:OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs

View a PDF of the paper titled OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs, by Guangzhi Sun and 3 other authors

View PDF HTML (experimental)

Abstract:Audio-visual large language models (LLMs) hold strong promise for long-form video understanding, yet their long-video inference is fundamentally limited by the linear growth of video tokens and key-value (KV) caches. We present OmniMem, a memory-efficient streaming framework designed specifically for audio-visual LLMs. Unlike existing compression methods that treat all tokens uniformly, OmniMem introduces a modality-aware memory allocation strategy that separately manages visual and audio contexts, addressing the severe token imbalance between the two modalities. OmniMem further preserves informative and non-redundant KV states through perturbation-aware memory selection, enabling compact memory without sacrificing long-range understanding. To strengthen compression under realistic deployment constraints, we also explore budget-aware fine-tuning, which encourages the model to consolidate useful information into retained memory. Experiments on VideoMME Long, LVBench, and LVOmniBench with video-SALMONN 2+ and Qwen-2.5-Omni show that OmniMem consistently improves over strong training-free compression baselines by 2-4% absolute accuracy under the same memory budgets, with an additional 1-2% gain after fine-tuning.

Comments: Code: this https URL

Subjects:

Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Sound (cs.SD); Audio and Speech Processing (eess.AS)

Cite as: arXiv:2606.07577 [cs.AI]

(or arXiv:2606.07577v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2606.07577

arXiv-issued DOI via DataCite

Submission history

From: Guangzhi Sun [view email] [v1] Tue, 26 May 2026 19:12:13 UTC (2,375 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled OmniMem: Perturbation-aware Memory Compression for Streaming Audio-Visual LLMs, by Guangzhi Sun and 3 other authors

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.AI

new | recent | 2026-06

Change to browse by:

cs cs.CV cs.SD eess eess.AS

References & Citations

NASA ADS

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Introduces modality-aware memory allocation that separately manages visual and audio contexts to address severe token imbalance between modalities.
  • Uses perturbation-aware KV state selection to retain informative and non-redundant memory while enabling compact compression.
  • Explores budget-aware fine-tuning to encourage the model to consolidate useful information into retained memory.
  • Achieves 2-4% absolute accuracy gains over training-free compression baselines on three long-video benchmarks with two audio-visual models, plus 1-2% after fine-tuning.

Highlights and analysis are generated automatically and may contain errors. Check the original source.