May 26, 2026 · Artificial Intelligence

When Should a Streaming Video LLM Speak? Evidence‑Condition Alignment via Explicit Scene Graphs (Response‑G1)

The ACL 2026 paper introduces Response‑G1, a proactive streaming video‑LLM framework that aligns visual evidence with response conditions using explicit scene‑graph modeling, memory‑augmented retrieval, and trigger‑based decision making, achieving 12.8 % and 15.1 % improvements on active tasks of OVO‑Bench and StreamingBench while also benefiting passive settings.

Proactive InteractionResponse-G1Scene Graph

0 likes · 9 min read

When Should a Streaming Video LLM Speak? Evidence‑Condition Alignment via Explicit Scene Graphs (Response‑G1)

Machine Heart

Apr 20, 2026 · Artificial Intelligence

AURA: Real-Time Video Understanding Shifts from Post-Play Q&A to Continuous Interaction

AURA introduces an always‑on video LLM that processes streams frame‑by‑frame, decides when to stay silent or answer, uses a dual sliding‑window context and a Silent‑Speech Balanced Loss, achieves state‑of‑the‑art scores on StreamingBench, OVO‑Bench and OmniMMI, and runs at 2 FPS with ~312 ms end‑to‑end latency on two 80G GPUs.

AURABenchmarkReal-Time Interaction

0 likes · 15 min read

AURA: Real-Time Video Understanding Shifts from Post-Play Q&A to Continuous Interaction