Research map / Video / Video-language understanding
X-Stream: Benchmarking MLLMs as Multiplexers for Multi-Stream Understanding
Peiwen Sun, Xudong LU, Huadai Liu, Yang Bo, Dongming Wu, Huankang Guan, Minghong Cai, Jinpeng Chen, Xintong Guo, Shuhan LI, FANG LIU, Rui Liu and 1 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 |
|---|---|---|---|
| use time division instead of spatial division multiplexing (ablation) | Gemini 3 Pro · X-Stream cross-stream interference | score: 66.96 -> 80.74 | Improved |
| use time division instead of spatial division multiplexing (ablation) | Gemini 3 Pro · X-Stream cross-stream reference | score: 74.79 -> 67.08 | Hurt |
| use time division instead of spatial division multiplexing (ablation) | Qwen3-Omni-30B-A3B · X-Stream cross-stream reference | score: 66.08 -> 58.62 | Hurt |
| use semantic instead of spatial division with 4 streams (ablation) | Gemini 3 Pro · X-Stream, 4 streams | score: 30.48 -> 41.25 | Improved |
| use semantic instead of spatial division with 5 streams (ablation) | Gemini 3 Pro · X-Stream, 5 streams | score: 22.81 -> 31.06 | Improved |
| use time instead of spatial division with 2 streams (ablation) | Qwen3-Omni-30B-A3B · X-Stream, 2 streams | score: 36.61 -> 40.55 | Improved |
| use semantic instead of spatial division on overall single-stream ability (ablation) | Qwen3-Omni-30B-A3B · X-Stream single stream ability | score: 63.41 -> 61.58 | Hurt |
| add audio input (ablation) | Gemini 3 Pro · X-Stream overall (spatial) | score: 46.13 -> 49.60 | Improved |
| add audio input with time division (ablation) | Qwen3-Omni-30B-A3B · X-Stream overall | score: 30.37 -> 26.57 (time); 29.40 -> 34.28 (spatial) | Mixed |
| inject distractor streams into single-stream benchmark (ablation) | Gemini 3 Pro · StreamingBench | score: 65.20 -> 35.91 | Hurt |
| give single stream input on multi-stream benchmark (ablation) | Gemini 3 Pro · X-Stream | score: 49.60 -> 13.93 | Hurt |
Closest papers by meaning
- OmniMMI: A Comprehensive Multi-modal Interaction Benchmark in Streaming Video Contexts (CVPR 2025 · similarity 0.88)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.86)
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.85)
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.85)
- StreamBridge: Turning Your Offline Video Large Language Model into a Proactive Streaming Assistant (NeurIPS 2025 · similarity 0.85)
- MVU-Eval: Towards Multi-Video Understanding Evaluation for Multimodal LLMs (NeurIPS 2025 · similarity 0.85)
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.
