Research map / Video / Video-language understanding
OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models
Yue Ding, Yiyan Ji, Jungang Li, Xuyang Liu, Xinlong Chen, Junfei Wu, Bozhou Li, Bohan Zeng, Yang Shi, Yushuo Guan, Yuanxing Zhang, Jiaheng Liu and 3 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add spatial saliency to STVP (vs temporal only) (ablation) | OmniSIFT · DailyOmni, WorldSense | accuracy | Improved |
| add temporal saliency to STVP (vs spatial only) (ablation) | OmniSIFT · DailyOmni, WorldSense | accuracy | Improved |
| add visual guidance to audio selection (vs audio-only self-attention selector) (ablation) | OmniSIFT VGAS · DailyOmni, WorldSense | accuracy: accuracy declines of 3.9% and 2.9% on DailyOmni and WorldSense | Improved |
| use 1-layer instead of 3-layer VGAS selector (ablation) | OmniSIFT · VideoMME, 35% | accuracy: 67.2 -> 68.3 | Improved |
| use 1-layer instead of 3-layer VGAS selector (ablation) | OmniSIFT · DailyOmni, 35% | accuracy: 72.3 -> 73.2 | Improved |
| use 1-layer instead of 3-layer VGAS selector (ablation) | OmniSIFT · WorldSense, 35% | accuracy: 49.0 -> 50.0 | Improved |
| use STVP instead of random video pruning (ablation) | OmniSIFT · DailyOmni, 35% | accuracy: 67.8 -> 73.2 | Improved |
| use STVP instead of random video pruning (ablation) | OmniSIFT · WorldSense, 35% | accuracy: 47.8 -> 50.0 | Improved |
| use VGAS instead of random audio pruning (ablation) | OmniSIFT · DailyOmni, 35% | accuracy: 71.0 -> 73.2 | Improved |
| use VGAS instead of random audio pruning (ablation) | OmniSIFT · WorldSense, 35% | accuracy: 49.1 -> 50.0 | Improved |
| use cross-attention VGAS instead of SSM selector (ablation) | OmniSIFT · DailyOmni, 35% | accuracy: 67.3 -> 73.2 | Improved |
| use fixed instead of adaptive video budget allocation (ablation) | OmniSIFT · DailyOmni, 35% | accuracy: 71.9 -> 73.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniZip — OmniZip: Audio-Guided Dynamic Token Compression for Fast Omnimodal Large Language Models (CVPR 2026) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OmniSelect: Dynamic Modality-Aware Token Compression for Efficient Omni-modal Large Language Models (NeurIPS 2026 · similarity 0.95)
- AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression (NeurIPS 2026 · similarity 0.90)
- One Token per Highly Selective Frame: Towards Extreme Compression for Long Video Understanding (NeurIPS 2025 · similarity 0.90)
- LongVALE: Vision-Audio-Language-Event Benchmark Towards Time-Aware Omni-Modal Perception of Long Videos (CVPR 2025 · similarity 0.89)
- OmniVideoBench: Towards Audio-Visual Understanding Evaluation for Omni MLLMs (ICLR 2026 · similarity 0.88)
- Omni-Interactive Universal Embedder (NeurIPS 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
