Reproduce “BLIP-Live: Balancing Timeliness & Firing Precision in Proactive Streaming Video Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Kanchana Ranasinghe, Honglu Zhou, Dominick Reilly, Le Xue, Shusheng Yang, Juan Carlos Niebles, Silvio Savarese, Ran Xu

BLIP-Live trains video assistants to decide whether to answer or stay silent as frames arrive. It uses human-checked response and silence labels, and a three-state system that responds, holds an answer, or remains silent; it evaluates quality, timeliness, and firing precision. On ProactiveVideoQA, its Qwen model’s mean PAUC-F is 49.08 versus MMDuet2’s 34.01 at ω=0.5, with harness setups as reported.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.