Reproduce “VideoPro: Adaptive Program Reasoning for Long 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

Chenglin Li, Feng Han, Yikun Wang, Ruilin Li, Shuai Dong, Haowen Hou, Haitao Li, Qianglong Chen and 4 more

VideoPro lets a trained VideoLLM answer long-video questions directly or generate executable programs that retrieve, localize, and inspect evidence, then revise them after execution failures or low-confidence answers. Across four benchmark settings, Qwen3-VL-8B with VideoPro averages 59.6% accuracy versus 52.9% natively. Refinement improves accuracy/runtime trade-offs, but adds runtime and confidence can be misleading.

Summary by alphaXiv.

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