Reproduce “SimplePro: A Simple Baseline for Proactive Video Understanding” 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
Yilin Wang, Xiangxi Zheng, Dongxing Mao, Linjie Li, Zhengyuan Yang, Alex Jinpeng Wang
SimplePro adapts a frozen vision-language model for proactive video responses using explicit reporting rules, recent frames, retained observations, and a record of prior answers. The model decides when to answer, while a delivery filter removes consecutive duplicates. With a frozen Qwen3-VL-4B backbone, it scores 54.80% on StreamingBench-PO Acc@2s, 13.60 percentage points above SyncPro-RL in the reported benchmark setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.