PhoStream: Benchmarking Real-World Streaming for Omnimodal Assistants in Mobile Scenarios
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914293517123584 |
|---|---|
| author | Lu, Xudong Guan, Huankang Bo, Yang Chen, Jinpeng Guo, Xintong Li, Shuhan Liu, Fang Sun, Peiwen Li, Xueying Zhang, Wei Yang, Xue Liu, Rui Li, Hongsheng |
| author_facet | Lu, Xudong Guan, Huankang Bo, Yang Chen, Jinpeng Guo, Xintong Li, Shuhan Liu, Fang Sun, Peiwen Li, Xueying Zhang, Wei Yang, Xue Liu, Rui Li, Hongsheng |
| contents | Multimodal Large Language Models excel at offline audio-visual understanding, but their ability to serve as mobile assistants in continuous real-world streams remains underexplored. In daily phone use, mobile assistants must track streaming audio-visual inputs and respond at the right time, yet existing benchmarks are often restricted to multiple-choice questions or use shorter videos. In this paper, we introduce PhoStream, the first mobile-centric streaming benchmark that unifies on-screen and off-screen scenarios to evaluate video, audio, and temporal reasoning. PhoStream contains 5,572 open-ended QA pairs from 578 videos across 4 scenarios and 10 capabilities. We build it with an Automated Generative Pipeline backed by rigorous human verification, and evaluate models using a realistic Online Inference Pipeline and LLM-as-a-Judge evaluation for open-ended responses. Experiments reveal a temporal asymmetry in LLM-judged scores (0-100): models perform well on Instant and Backward tasks (Gemini 3 Pro exceeds 80), but drop sharply on Forward tasks (16.40), largely due to early responses before the required visual and audio cues appear. This highlights a fundamental limitation: current MLLMs struggle to decide when to speak, not just what to say. Code and datasets used in this work will be made publicly accessible at https://github.com/Lucky-Lance/PhoStream. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_22575 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | PhoStream: Benchmarking Real-World Streaming for Omnimodal Assistants in Mobile Scenarios Lu, Xudong Guan, Huankang Bo, Yang Chen, Jinpeng Guo, Xintong Li, Shuhan Liu, Fang Sun, Peiwen Li, Xueying Zhang, Wei Yang, Xue Liu, Rui Li, Hongsheng Computer Vision and Pattern Recognition Computation and Language Multimodal Large Language Models excel at offline audio-visual understanding, but their ability to serve as mobile assistants in continuous real-world streams remains underexplored. In daily phone use, mobile assistants must track streaming audio-visual inputs and respond at the right time, yet existing benchmarks are often restricted to multiple-choice questions or use shorter videos. In this paper, we introduce PhoStream, the first mobile-centric streaming benchmark that unifies on-screen and off-screen scenarios to evaluate video, audio, and temporal reasoning. PhoStream contains 5,572 open-ended QA pairs from 578 videos across 4 scenarios and 10 capabilities. We build it with an Automated Generative Pipeline backed by rigorous human verification, and evaluate models using a realistic Online Inference Pipeline and LLM-as-a-Judge evaluation for open-ended responses. Experiments reveal a temporal asymmetry in LLM-judged scores (0-100): models perform well on Instant and Backward tasks (Gemini 3 Pro exceeds 80), but drop sharply on Forward tasks (16.40), largely due to early responses before the required visual and audio cues appear. This highlights a fundamental limitation: current MLLMs struggle to decide when to speak, not just what to say. Code and datasets used in this work will be made publicly accessible at https://github.com/Lucky-Lance/PhoStream. |
| title | PhoStream: Benchmarking Real-World Streaming for Omnimodal Assistants in Mobile Scenarios |
| topic | Computer Vision and Pattern Recognition Computation and Language |
| url | https://arxiv.org/abs/2601.22575 |