新闻
Nunchi: Ears Help Eyes in Multi-Party Conversations
arXiv cs.AI · 发布于 · 阅读约3分钟
30秒读懂
- 发生了什么
- Nunchi framework improves gaze detection in multi-party conversations by combining visual gaze geometry with speech and transcript cues through vision-language models.
- 为何重要
- Matters for engineers building video understanding systems, meeting analysis tools, or attention-tracking applications that must identify who someone is looking at.
- 注意
- Requires synchronized speech, transcripts, and speaker identification; gains depend on transcript quality and may not generalize to scenarios lacking reliable audio or speech data.
- speech
- benchmark
这条新闻背后的模式
每个模式都讲清楚技术如何运作、何时值得投入,以及在哪里会失效。
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。