Reproduce “VideoACE: Benchmarking Agentic Video Understanding through High-Frame-Rate Actions” 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
Luoyuan Zhang, Xuanyu Zheng, Yancheng Long, Haoyang Xu, Bin Wen, Tingting Gao
VideoACE evaluates whether models can use interaction to gather evidence about fast actions. It contains 1,000 four-choice questions across 100 silent videos, with annotations marking brief relevant moments. Models can request denser clips or selected frames and regions. Gemini 3.6 Flash improves from 65.6% to 79.8% with tool use, a 14.2-point gain on this benchmark, not a general guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.