Reproduce “Event-STVG: Spatio-Temporal Video Grounding for Long-Form Event 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
Yizhe Zhang, Xujia Wang, Songqi Tang, Xujie Tang, Leo Jiahao Jiang, Jiacheng Wang, Bowen Qu, JinBo Li and 5 more
Event-STVG introduces a benchmark requiring models to locate a specified event occurrence in video and return its time interval plus box trajectories for requested participant roles. Queries specify event type, ordinal occurrence, and semantic roles, rather than participants’ appearance. On a fixed 300-query soccer subset at 10 minutes, Gemini-3.7 Flash achieves 5.40% m-vIoU, a joint grounding score, not a full-benchmark result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.