Reproduce “Think in 360°: Benchmarking and Bootstrapping Panoramic Video Reasoning” 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

Lidong Lu, Lidong Zou, Xinyi Mao, Jiaxin Lv, Zhenquan Li, Liangrui Dong, Tong Lu, Guo Chen

PanoVideoReasoner trains a video model to connect verified observations across directions and time in panoramic video. It uses annotated questions and reasoning traces, with evidence views selected by timestamp and direction to guide training. On PanoVideoBench, it reaches 55.7% overall accuracy, versus 43.2% for its Qwen3.5-9B backbone, but this result is limited to the benchmark’s evaluation protocol.

Summary by alphaXiv.

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