StreamVibe-VL infers evolving user intent for proactive video agents
Original title: StreamVibe: Proactive Video Agents for Evolving User Intent
The benchmark has 4.1K evaluation and 85K training samples; timing and answer quality rose 10.7% over baselines.
Researchers introduce StreamVibe-Bench for explicit, implicit, and dynamic user intents, plus StreamVibe-VL, a streaming model that infers intent from continuous observations. StreamVibe-Harness keeps intents as persistent states with hierarchical memory. The model retained general video understanding while improving response timing and answer quality. Existing proactive models remained weak on implicit and dynamic intents.
- StreamVibe-Bench contains 4.1K evaluation samples and 85K training samples.
- Response timing and answer quality improved 10.7% over other baselines.
- StreamVibe-Harness added a 12.3% gain in proactive interaction.
- Prior proactive models struggled to decide when to respond.
From the original article
1 min read at huggingface.co
Vision-language models are shifting from offline passive understanding toward online proactive interaction. However, prevailing methods largely rely on explicit requests while miss the implicit user intent (e. g.
By Chengzhi Liu, Yuxiang Lai, Yuzhe Yang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.