Research map / Vision-language and multimodal / Cross-modal learning
PRIM:Cooperative Dynamic Token Compression for Efficient Large Multimodal Models
Song Li, yongping xiong
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 |
|---|---|---|---|
| apply PRIM compression at 50% token retention instead of full tokens | Qwen2.5-Omni-7B · Video benchmarks average | average score: 61.4 -> 60.4 | No clear effect |
| use PRIM instead of FastVID at 50% retention | Qwen2.5-Omni-7B · AVUT average | accuracy: 60.2 -> 62.3 | Improved |
| use PRIM instead of FastVID at 35% retention | Qwen2.5-Omni-7B · Video benchmarks average | average score: 55.7 -> 56.7 | Improved |
| use PRIM instead of FastVID on long videos at 50% retention | Qwen2.5-Omni-7B · VideoMME long | accuracy: 49.4 -> 48.9 | Hurt |
| use PRIM instead of FastVID on LLaVA-Video at 25% retention | LLaVA-Video · VideoMME overall | accuracy: 59.3 -> 61.8 | Improved |
| use PRIM instead of FastVID on LLaVA-OV-7B at 50% retention | LLaVA-OV-7B · WorldSense average | accuracy: 50.4 -> 51.0 | Improved |
| use PRIM instead of FastVID at 50% retention on Qwen2.5-Omni-7B | Qwen2.5-Omni-7B · WorldSense latency per example | latency (s): 3.64s -> 3.29s | Improved |
| use PRIM instead of FastVID on Qwen2.5-Omni-3B AVUT event localization | Qwen2.5-Omni-3B · AVUT EL | accuracy: 42.1 -> 43.7 | Improved |
| use PRIM instead of FastVID on counting subtask of AVUT at 35% | Qwen2.5-Omni-3B · AVUT CC | accuracy: 39.6 -> 38.7 | Hurt |
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.
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 3 of 3 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 0.89)
- Task-Related Token Compression in Multimodal Large Language Models from an Explainability Perspective (ICLR 2026 · similarity 0.88)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.88)
- JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation (ICLR 2026 · similarity 0.87)
- AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression (NeurIPS 2026 · similarity 0.87)
- LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning (NeurIPS 2026 · similarity 0.87)
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.
